Last 7 Days (September 09 – September 15, 2026)
We study how model post-training and test-time inference design affect natural-language proof generation for hard olympiad mathematics. Starting from Nemotron 3 Ultra, we train two specialist checkpoints using supervised fine-tuning and reinforcement learning, and evaluate checkpoint choice, verification, and refinement. Based on these findings, we present an open-model test-time-compute pipeline. The system operates entirely in natural language, with no formal prover, external tools, or internet access. Three Nemotron 3 Ultra checkpoints - the general-availability model and two post-trained specialists - power an iterative search that generates, verifies, and refines candidate proofs; a separate high-compute stage then selects each final submission. The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. We release the two post-trained checkpoints as well as the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.
Primary: NVIDIA
All Institutions: NVIDIA
The paper presents an open-model test-time-compute pipeline that achieves IMO Gold medal performance using natural language proofs. It contributes a rigorous methodology for iterative proof generation and verification, along with a new benchmark and open resources, marking a significant step in the capabilities of open-source LLMs for complex reasoning tasks.
The paper proposes a sophisticated test-time compute pipeline for natural language proof generation. The core methodology involves training two specialist checkpoints from Nemotron 3 Ultra using SFT and RL, then orchestrating an iterative search process that generates, verifies, and refines candidate proofs without relying on formal provers or external tools. The inclusion of a high-compute selection stage for final submissions is a key architectural decision that differentiates this from simple best-of-N sampling.
The primary result is the achievement of 30/42 points at IMO 2026, reaching the gold-medal threshold. This is a significant empirical milestone for open models. The introduction of Nemotron-IMO-Bench (200 novel problems) provides a necessary evaluation framework, though the sample size is relatively small compared to standard benchmarks. The ablation studies on checkpoint choice and verification strategies are critical to understanding the pipeline's effectiveness.
The paper explicitly states the release of post-trained checkpoints, training data, code, and submitted solutions. This high level of transparency significantly enhances reproducibility, allowing the community to verify the results and build upon the pipeline.
The reliance on a specific base model (Nemotron 3 Ultra) may limit immediate generalizability to other architectures. The benchmark size (200 problems) is small for statistical robustness. The high compute cost of the iterative search and selection stages may be prohibitive for some users.
This work demonstrates that open models can achieve human-competitive performance in high-level mathematical reasoning, potentially accelerating research in automated theorem proving and educational AI. The open release of assets fosters community development in this niche. The paper presents an open-model test-time-compute pipeline that achieves IMO Gold medal performance using natural language proofs. It contributes a rigorous methodology for iterative proof generation and verification, along with a new benchmark and open resources, marking a significant step in the capabilities of open-source LLMs for complex reasoning tasks.
Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution. Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors.
Primary: National University of Singapore (MagicLab)
All Institutions: National University of Singapore
The paper introduces Latent Interface Training (LIT), a two-stage strategy that mitigates vision-action shortcuts in robot foundation models by decoupling goal-directed action learning from visual conditioning and supervising a latent interface with terminal pose. This approach demonstrates significant improvements in generalization across multiple architectures and real-world distribution shifts, offering a practical and effective solution to a critical challenge in deploying robotic foundation models.
The paper proposes Latent Interface Training (LIT), a two-stage framework designed to mitigate "vision-action shortcuts" in robot foundation models. The core idea is to decouple the learning of goal-directed action generation from visual conditioning. Stage 1 trains an action expert to generate actions based on language, robot state, and terminal SE(3) pose, effectively creating a spatial-goal-conditioned prior without relying on images. Stage 2 introduces a latent interface that aggregates visual and semantic representations, supervised to reconstruct the terminal pose used in Stage 1. This forces the visual pathway to retain only goal-relevant spatial information, filtering out task-irrelevant visual cues. The method is framework-agnostic, applied to Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM. The methodology is sound and addresses a known issue in robotics foundation models (over-reliance on spurious visual correlations).
The paper evaluates LIT across four different architectures on the LIBERO-Plus benchmark, reporting improvements of 3.87-10.70 percentage points in success rates while maintaining or improving average LIBERO success. Real-world evaluations show significant gains (13.30-16.70 percentage points) under distribution shifts such as unseen camera configurations, lighting variations, and distractors. The breadth of architectures tested is a strength, demonstrating the generalizability of the training strategy. However, the specific details of the real-world setup and the magnitude of the baseline performance are not fully detailed in the provided text, making it hard to assess the absolute difficulty of the tasks.
The paper provides a project page link. The method is described as a training strategy rather than a complex new architecture, which generally aids reproducibility. However, without access to the full code and specific hyperparameters for the "latent interface" and the exact implementation details of the pose supervision, full reproducibility is difficult to guarantee. The use of standard benchmarks like LIBERO helps.
The primary limitation is the reliance on terminal SE(3) pose as the supervisory signal for the latent interface. This assumes that the terminal pose is a sufficient proxy for the goal-relevant spatial information needed for all action chunks, which may not hold for highly complex or long-horizon tasks where intermediate spatial states are critical. Additionally, the method adds a two-stage training process, which may increase computational overhead compared to single-stage training.
This work has significant potential for the robotics community, as improving generalization under visual distribution shifts is a major bottleneck for deploying foundation models in real-world environments. By providing a framework-agnostic training strategy, it can be applied to a wide range of existing VLA models. The insights into mitigating shortcut learning in multimodal robotics are valuable for the broader field of embodied AI. The paper introduces Latent Interface Training (LIT), a two-stage strategy that mitigates vision-action shortcuts in robot foundation models by decoupling goal-directed action learning from visual conditioning and supervising a latent interface with terminal pose. This approach demonstrates significant improvements in generalization across multiple architectures and real-world distribution shifts, offering a practical and effective solution to a critical challenge in deploying robotic foundation models.
Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool-call turns per task, rewarded by executing each task's own verifier. We provide a comprehensive recipe: First, an aggressively warm-started to stabilize actor-critic training, with a dense process reward scoring trajectories by the absolute number of passing verifiers. Second, stable optimization through TITO construction, training on the exact sampled token identifiers with drift repair at turn boundaries, and rollout routing replay, recording the sampler's per-token expert choices at every MoE layer and replaying them during training. Third, fully out-of-distribution training corpus: isolated seeds and synthesized tasks disjoint from Terminal-Bench 2.1 ensures gains reflect genuine capability transfer over benchmark overfitting. Together, TITO and R3 cut the training-to-inference log-probability difference from 0.021 to 0.013, with exactly aligned zero token drift in the loss region. On Terminal-Bench 2.1, our post-train pipeline raises initial base model from 43.8% to T1 with 64.0% resolved. On Long-Horizon Terminal Bench, T1 reaches 27.9% and surpasses GPT-5.4 and GLM-5.1.
Primary: Alibaba Group
All Institutions: Alibaba Group, Tsinghua University
The paper presents a comprehensive and technically rigorous approach to training large MoE agents via RL, introducing novel stabilization techniques (TITO and R3) that are critical for scaling RL to sparse models. Its strong empirical results on terminal benchmarks demonstrate a significant leap in autonomous software engineering capabilities, establishing a new standard for long-horizon agentic tasks.
The paper introduces a robust framework for training large Mixture-of-Experts (MoE) models using Reinforcement Learning (RL) in long-horizon, stateful environments (Linux terminals). The core technical contribution is the "TITO" (Token-In-Token-Out) and "R3" (Rollout Routing Replay) mechanisms. TITO addresses the training-inference mismatch caused by tokenization drift in multi-turn agentic loops by ensuring the trainer consumes the exact token identifiers generated by the sampler, repairing boundary discrepancies. R3 addresses the unique challenge of MoE models where discrete expert routing decisions made during inference can differ from those recomputed during training due to numerical precision differences; R3 records the expert masks during rollout and replays them during the training forward pass. Additionally, the paper proposes a dense process reward based on the absolute number of passing verifier assertions, which provides a much richer learning signal than binary success/fail rewards for difficult tasks. The methodology is rigorous, addressing specific, non-trivial engineering challenges in scaling RL to sparse models.
The experiments are extensive, utilizing a 122B parameter MoE model (Qwen3.5-122B-A10B) trained on 15,000 synthesized terminal tasks. The evaluation is conducted on three held-out benchmarks: Terminal-Bench 2.1, Long-Horizon Terminal-Bench (LHTB), and Terminal-Bench Hard. The results show a significant improvement from the base model (43.8%) to the final T1 model (64.0%) on Terminal-Bench 2.1, surpassing several frontier models like GPT-5.4 and GLM-5.1. The ablation studies clearly demonstrate the necessity of both TITO and R3 for stabilizing training, as well as the benefit of the dense reward over binary rewards. The analysis of training dynamics, including explained variance and turn count growth, provides deep insight into the learning process.
The paper provides high levels of detail regarding the training infrastructure, hyperparameters, and specific mechanisms like the routing replay algorithm. It describes the dataset construction process (RST) and the audit criteria for task selection. However, the specific code for the TITO and R3 implementations is not explicitly linked in the provided text (though likely available via the authors' repositories), and the exact hardware configuration for the 122B model training is only partially detailed. The use of proprietary or specific internal tools (like the "slime" framework) may limit immediate reproducibility for external labs without access to those specific infrastructure components.
The primary limitation is the reliance on a large-scale, proprietary infrastructure to train a 122B MoE model, making the results difficult to replicate for smaller labs. The evaluation is limited to terminal tasks, and while the paper argues for generalization, the specific domain of Linux shell operations is narrow. Additionally, the paper acknowledges that the model still struggles with the hardest tasks (Terminal-Bench Hard) compared to some larger dense models, and the "inefficient search" behavior (high turn counts) suggests that the model has not fully mastered optimal planning, merely improving its persistence and partial success rates.
This work is significant for the development of autonomous agents capable of performing complex, long-horizon software engineering tasks. By solving the stability issues of RL in MoE models, it paves the way for more efficient and capable agentic systems. The dense reward design is a valuable contribution to the RL community, offering a practical solution to the sparse reward problem in verifiable environments. The findings on training-inference consistency in sparse models will be relevant to any group working on large-scale MoE RL. The paper presents a comprehensive and technically rigorous approach to training large MoE agents via RL, introducing novel stabilization techniques (TITO and R3) that are critical for scaling RL to sparse models. Its strong empirical results on terminal benchmarks demonstrate a significant leap in autonomous software engineering capabilities, establishing a new standard for long-horizon agentic tasks.
Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution. Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors.
Primary: National University of Singapore (MagicLab)
All Institutions: National University of Singapore
The paper introduces Latent Interface Training (LIT), a two-stage strategy that mitigates vision-action shortcuts in robot foundation models by decoupling goal-directed action learning from visual conditioning and supervising a latent interface with terminal pose. This approach demonstrates significant improvements in generalization across multiple architectures and real-world distribution shifts, offering a practical and effective solution to a critical challenge in deploying robotic foundation models.
The paper proposes Latent Interface Training (LIT), a two-stage framework designed to mitigate "vision-action shortcuts" in robot foundation models. The core idea is to decouple the learning of goal-directed action generation from visual conditioning. Stage 1 trains an action expert to generate actions based on language, robot state, and terminal SE(3) pose, effectively creating a spatial-goal-conditioned prior without relying on images. Stage 2 introduces a latent interface that aggregates visual and semantic representations, supervised to reconstruct the terminal pose used in Stage 1. This forces the visual pathway to retain only goal-relevant spatial information, filtering out task-irrelevant visual cues. The method is framework-agnostic, applied to Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM. The methodology is sound and addresses a known issue in robotics foundation models (over-reliance on spurious visual correlations).
The paper evaluates LIT across four different architectures on the LIBERO-Plus benchmark, reporting improvements of 3.87-10.70 percentage points in success rates while maintaining or improving average LIBERO success. Real-world evaluations show significant gains (13.30-16.70 percentage points) under distribution shifts such as unseen camera configurations, lighting variations, and distractors. The breadth of architectures tested is a strength, demonstrating the generalizability of the training strategy. However, the specific details of the real-world setup and the magnitude of the baseline performance are not fully detailed in the provided text, making it hard to assess the absolute difficulty of the tasks.
The paper provides a project page link. The method is described as a training strategy rather than a complex new architecture, which generally aids reproducibility. However, without access to the full code and specific hyperparameters for the "latent interface" and the exact implementation details of the pose supervision, full reproducibility is difficult to guarantee. The use of standard benchmarks like LIBERO helps.
The primary limitation is the reliance on terminal SE(3) pose as the supervisory signal for the latent interface. This assumes that the terminal pose is a sufficient proxy for the goal-relevant spatial information needed for all action chunks, which may not hold for highly complex or long-horizon tasks where intermediate spatial states are critical. Additionally, the method adds a two-stage training process, which may increase computational overhead compared to single-stage training.
This work has significant potential for the robotics community, as improving generalization under visual distribution shifts is a major bottleneck for deploying foundation models in real-world environments. By providing a framework-agnostic training strategy, it can be applied to a wide range of existing VLA models. The insights into mitigating shortcut learning in multimodal robotics are valuable for the broader field of embodied AI. The paper introduces Latent Interface Training (LIT), a two-stage strategy that mitigates vision-action shortcuts in robot foundation models by decoupling goal-directed action learning from visual conditioning and supervising a latent interface with terminal pose. This approach demonstrates significant improvements in generalization across multiple architectures and real-world distribution shifts, offering a practical and effective solution to a critical challenge in deploying robotic foundation models.
We introduce StepAudio 3 Gen, a general-purpose audio generation model that supports zero-shot text-to-speech (TTS), voice design, vocal generation, sound effects, music, vibe speech, and mixtures of multiple audio types within a unified framework. At its core, StepAudio 3 Gen is a discrete autoregressive generator that models audio directly over residual vector quantization (RVQ) tokens, departing from the diffusion Transformer-based continuous generation paradigm prevalent in recent general audio models. Its StepAudio Tokenizer represents general audio at 12.5 Hz in a shared 16 times 2048 residual code space, jointly quantizing semantic and waveform-level acoustic features so that each code layer preserves both types of information. For generation, the backbone predicts the first codebook along the time axis using autoregressive modeling, while a lightweight causal Transformer completes the remaining fifteen codebooks along the codebook axis. Our study further identifies three key design principles: (1) interference-aware progressive pretraining for acquiring audio capabilities while preserving the textual abilities of the large language model, (2) RVQ Adaptor for effectively incorporating multi-codebook acoustic representations, and (3) discrete autoregressive modeling over a shared representation across general audio domains. With progressive pretraining, multi-task instruction training, and supervised fine-tuning, StepAudio 3 Gen achieves state-of-the-art performance on both TTS and voice design, while retaining strong generation capabilities across speech, vocals, sound effects, and music. Audio samples are available at https://stepaudiollm.github.io/step-audio-3-gen/.
Primary: StepFun
All Institutions: StepFun
StepAudio 3 Gen introduces a unified discrete autoregressive framework for general audio generation that claims state-of-the-art performance across speech, music, and sound effects by leveraging a shared residual vector quantization tokenizer and a hybrid autoregressive architecture. The paper presents a significant architectural departure from diffusion-based models, offering a scalable path for integrating diverse audio tasks into large language models, though its impact is tempered by the proprietary nature of the implementation and the need for further independent verification of its claimed superiority over specialized baselines.
The paper proposes a unified discrete autoregressive framework for general audio generation, diverging from the dominant diffusion-based continuous generation paradigm. The core innovation lies in the "StepAudio Tokenizer," which operates at 12.5 Hz using a shared 16x2048 residual code space to jointly quantize semantic and acoustic features. The architecture employs a hybrid autoregressive strategy: the backbone LLM predicts the first codebook layer along the time axis, while a lightweight causal Transformer completes the remaining 15 layers along the codebook axis. This design aims to leverage the strong sequential modeling capabilities of LLMs for high-level structure while efficiently handling the high-dimensional acoustic details. The inclusion of an "RVQ Adaptor" and "interference-aware progressive pretraining" are presented as key techniques to integrate multi-codebook representations without degrading the base LLM's textual capabilities.
The paper claims state-of-the-art performance in TTS and voice design, as well as strong capabilities in vocal, sound effect, and music generation. However, the provided text is a technical report summary rather than a full peer-reviewed article with extensive ablation studies. While the scope of tasks (zero-shot TTS, voice design, music, SFX) is impressive and covers a wide range of audio domains, the evaluation section in the provided text is brief. It lacks detailed quantitative comparisons against specific baselines (e.g., VALL-E, MusicGen, or other recent unified models) in the text provided, relying heavily on the claim of SOTA performance. The "vibe speech" and mixture generation capabilities are novel but require rigorous human evaluation to validate quality, which is not fully detailed in the excerpt.
The paper provides a demo URL for audio samples, which is a positive step for qualitative assessment. However, as a technical report from a commercial entity (StepFun), the availability of code, model weights, and detailed hyperparameters is not explicitly confirmed in the text. The specific implementation details of the "interference-aware progressive pretraining" and the "RVQ Adaptor" are described conceptually but may lack the granular detail necessary for full independent reproduction without access to the proprietary codebase.
The primary limitation is the reliance on a proprietary tokenizer and training pipeline, which may limit community adoption compared to open-source alternatives. The 12.5 Hz frame rate is relatively low for high-fidelity audio reconstruction, potentially capping the maximum audio quality compared to higher-rate models. Furthermore, the "general-purpose" claim is ambitious; unifying music, speech, and SFX in a single discrete autoregressive model often leads to trade-offs in quality for individual domains compared to specialized models. The lack of detailed ablation studies in the provided text makes it difficult to isolate the contribution of each design principle.
This work contributes to the trend of unifying diverse audio generation tasks into single LLM-based frameworks. If the claimed SOTA performance is verified, it could shift the field's focus from diffusion models to discrete autoregressive models for audio, potentially offering faster inference and better integration with existing LLM ecosystems. The ability to generate mixtures of audio types is a significant step toward more flexible and creative audio production tools. StepAudio 3 Gen introduces a unified discrete autoregressive framework for general audio generation that claims state-of-the-art performance across speech, music, and sound effects by leveraging a shared residual vector quantization tokenizer and a hybrid autoregressive architecture. The paper presents a significant architectural departure from diffusion-based models, offering a scalable path for integrating diverse audio tasks into large language models, though its impact is tempered by the proprietary nature of the implementation and the need for further independent verification of its claimed superiority over specialized baselines.
Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool-call turns per task, rewarded by executing each task's own verifier. We provide a comprehensive recipe: First, an aggressively warm-started to stabilize actor-critic training, with a dense process reward scoring trajectories by the absolute number of passing verifiers. Second, stable optimization through TITO construction, training on the exact sampled token identifiers with drift repair at turn boundaries, and rollout routing replay, recording the sampler's per-token expert choices at every MoE layer and replaying them during training. Third, fully out-of-distribution training corpus: isolated seeds and synthesized tasks disjoint from Terminal-Bench 2.1 ensures gains reflect genuine capability transfer over benchmark overfitting. Together, TITO and R3 cut the training-to-inference log-probability difference from 0.021 to 0.013, with exactly aligned zero token drift in the loss region. On Terminal-Bench 2.1, our post-train pipeline raises initial base model from 43.8% to T1 with 64.0% resolved. On Long-Horizon Terminal Bench, T1 reaches 27.9% and surpasses GPT-5.4 and GLM-5.1.
Primary: Alibaba Group
All Institutions: Alibaba Group, Tsinghua University
The paper presents a comprehensive and technically rigorous approach to training large MoE agents via RL, introducing novel stabilization techniques (TITO and R3) that are critical for scaling RL to sparse models. Its strong empirical results on terminal benchmarks demonstrate a significant leap in autonomous software engineering capabilities, establishing a new standard for long-horizon agentic tasks.
The paper introduces a robust framework for training large Mixture-of-Experts (MoE) models using Reinforcement Learning (RL) in long-horizon, stateful environments (Linux terminals). The core technical contribution is the "TITO" (Token-In-Token-Out) and "R3" (Rollout Routing Replay) mechanisms. TITO addresses the training-inference mismatch caused by tokenization drift in multi-turn agentic loops by ensuring the trainer consumes the exact token identifiers generated by the sampler, repairing boundary discrepancies. R3 addresses the unique challenge of MoE models where discrete expert routing decisions made during inference can differ from those recomputed during training due to numerical precision differences; R3 records the expert masks during rollout and replays them during the training forward pass. Additionally, the paper proposes a dense process reward based on the absolute number of passing verifier assertions, which provides a much richer learning signal than binary success/fail rewards for difficult tasks. The methodology is rigorous, addressing specific, non-trivial engineering challenges in scaling RL to sparse models.
The experiments are extensive, utilizing a 122B parameter MoE model (Qwen3.5-122B-A10B) trained on 15,000 synthesized terminal tasks. The evaluation is conducted on three held-out benchmarks: Terminal-Bench 2.1, Long-Horizon Terminal-Bench (LHTB), and Terminal-Bench Hard. The results show a significant improvement from the base model (43.8%) to the final T1 model (64.0%) on Terminal-Bench 2.1, surpassing several frontier models like GPT-5.4 and GLM-5.1. The ablation studies clearly demonstrate the necessity of both TITO and R3 for stabilizing training, as well as the benefit of the dense reward over binary rewards. The analysis of training dynamics, including explained variance and turn count growth, provides deep insight into the learning process.
The paper provides high levels of detail regarding the training infrastructure, hyperparameters, and specific mechanisms like the routing replay algorithm. It describes the dataset construction process (RST) and the audit criteria for task selection. However, the specific code for the TITO and R3 implementations is not explicitly linked in the provided text (though likely available via the authors' repositories), and the exact hardware configuration for the 122B model training is only partially detailed. The use of proprietary or specific internal tools (like the "slime" framework) may limit immediate reproducibility for external labs without access to those specific infrastructure components.
The primary limitation is the reliance on a large-scale, proprietary infrastructure to train a 122B MoE model, making the results difficult to replicate for smaller labs. The evaluation is limited to terminal tasks, and while the paper argues for generalization, the specific domain of Linux shell operations is narrow. Additionally, the paper acknowledges that the model still struggles with the hardest tasks (Terminal-Bench Hard) compared to some larger dense models, and the "inefficient search" behavior (high turn counts) suggests that the model has not fully mastered optimal planning, merely improving its persistence and partial success rates.
This work is significant for the development of autonomous agents capable of performing complex, long-horizon software engineering tasks. By solving the stability issues of RL in MoE models, it paves the way for more efficient and capable agentic systems. The dense reward design is a valuable contribution to the RL community, offering a practical solution to the sparse reward problem in verifiable environments. The findings on training-inference consistency in sparse models will be relevant to any group working on large-scale MoE RL. The paper presents a comprehensive and technically rigorous approach to training large MoE agents via RL, introducing novel stabilization techniques (TITO and R3) that are critical for scaling RL to sparse models. Its strong empirical results on terminal benchmarks demonstrate a significant leap in autonomous software engineering capabilities, establishing a new standard for long-horizon agentic tasks.
Commercial text-to-image systems silently revise user prompts before generating images, a step users typically cannot disable or even see. Yet, existing audits of cultural bias examine only the final images and treat generation as a single pipeline, so they cannot tell where the bias originates. We introduce WORLDVIEW, a multilingual benchmark of 8,960 prompts across 15 languages and 31 language-context pairings. Using it, we audit the revision layer in three systems (DALL-E-3, Imagen-4, GPT-Image-1.5) through a three-step analysis of how heavily it marks each cultural context, whether it flattens that context into a narrow vocabulary, and whether that vocabulary is stereotypical. Relative to a no-context English baseline, the US is the least-marked context, while non-Western and non-Anglophone contexts are marked far more heavily, flattened into narrow vocabularies applied across topically diverse prompts, and reduced to recognizable cultural stereotypes. Comparing images from original versus revised prompts on models without a revision layer, we identify the layer itself as a previously undocumented, causal source of this stereotyping. To locate cultural bias, and fix it, we must audit the system as deployed, not the model alone.
Primary: University of Zurich
All Institutions: University of Zurich, Stanford University
The paper identifies the prompt revision layer in commercial text-to-image systems as a previously undocumented, causal source of cultural bias and stereotyping. By introducing the WORLDVIEW benchmark and a three-step audit framework, it demonstrates that non-Western cultures are disproportionately flattened into narrow, stereotypical vocabularies before image generation, necessitating a shift in bias mitigation strategies to include the entire deployed pipeline.
The paper introduces a rigorous three-step analytical framework (Contextual Markedness Score, Cultural Flattening Score, and Stereotypical Content Analysis) to audit the "prompt revision" layer in commercial text-to-image systems. By isolating this intermediate text-to-text transformation, the authors address a critical blind spot in prior bias audits that treated the generation pipeline as monolithic. The methodology is sound, utilizing sentence embeddings for markedness and TF-IDF for flattening, with a well-designed causal ablation study using open-source models to verify that the revision layer, not just the image model, drives stereotypical outputs.
The experimental setup is robust, featuring a new multilingual benchmark (WORLDVIEW) with 8,960 prompts across 15 languages and 31 contexts. The evaluation covers three major commercial systems (DALL-E-3, Imagen-4, GPT-Image-1.5). The results clearly demonstrate that non-Western contexts are disproportionately marked and flattened into stereotypical vocabularies (e.g., Egypt reduced to pyramids, Finland to snow). The causal link established via ablation on SDXL and Flux models is a strong technical contribution, proving the revision layer is a distinct source of bias.
High. The authors release the benchmark, revised prompts, metrics, and evaluation toolkit on GitHub. The use of standard APIs and open-source models for the ablation study ensures that the findings can be replicated by other researchers.
The causal ablation is restricted to English-speaking contexts due to language confounds, though a case study on Switzerland partially mitigates this. The study relies on VQA models for image description, which may introduce their own biases, although the authors argue the comparative design mitigates this. The benchmark, while broad, still underrepresents certain regions like Sub-Saharan Africa.
This paper has significant implications for AI safety and fairness research. It shifts the focus of bias mitigation from just the image generation model to the entire deployed system, including preprocessing layers. This insight is crucial for developers of commercial T2I systems, as it highlights that debiasing the image model alone is insufficient if the prompt revision layer injects stereotypes. It provides a actionable framework for auditing these hidden layers. The paper identifies the prompt revision layer in commercial text-to-image systems as a previously undocumented, causal source of cultural bias and stereotyping. By introducing the WORLDVIEW benchmark and a three-step audit framework, it demonstrates that non-Western cultures are disproportionately flattened into narrow, stereotypical vocabularies before image generation, necessitating a shift in bias mitigation strategies to include the entire deployed pipeline.
Most of mathematical knowledge has been communicated through so-called informal use of mathematics and natural language. With large language models (LLMs) being highly adept in using natural language, they achieve strong performance, yet not perfect, in informal mathematical reasoning. Restraining LLMs to informal reasoning misses out on the opportunity to use the discrete verification abilities that machines offer through machine-checkable proofs. In this paper, we bridge the gap between informal and formal reasoning by integrating Lean signals into the informal reasoning process. We introduce Magenta, a training-free agentic pipeline that, given only a natural-language problem, produces an answer, expresses it as a Lean 4 statement, and constructs a machine-checked proof. A statement judge verifies whether the formalisation preserves the original problem, while an error-attribution judge routes failed attempts either to mathematical re-derivation or local Lean repair. Magenta achieves 100% accuracy across all evaluated olympiad benchmarks, including AIME 2025, AIME 2026, and HMMT February 2026. When paired with the open-weight K2-Horizon-7B reasoner, it solves all six IMO 2026 problems. Our analysis shows that statement adjudication is essential for preventing false certificates and that feedback-guided correction outperforms independent resampling on difficult problems.
Primary: Imperial College London
All Institutions: Imperial College London, University College London
Magenta introduces a training-free agentic pipeline that integrates Lean 4 formal verification into LLM mathematical reasoning, achieving 100% accuracy on AIME/HMMT benchmarks and solving all IMO 2026 problems with a 7B model by using statement adjudication and error attribution to guide self-correction.
The paper introduces "Magenta," a training-free agentic pipeline that integrates informal LLM reasoning with formal Lean 4 verification. The core methodological contribution is the closed-loop feedback mechanism: a "Statement Judge" ensures the formalization is faithful to the natural language problem (addressing the autoformalization gap), and an "Error Attribution Judge" routes verification failures to either mathematical re-derivation or local proof repair. This distinction is crucial, as it prevents the system from wasting compute on syntax errors when the math is wrong, or vice versa. The use of a deterministic verifier (Lean) as a hard constraint on LLM outputs is a strong architectural choice that differentiates it from purely learned reward models.
The experiments are extensive, covering AIME 2025/2026, HMMT 2026, and IMO 2026. The claim of 100% accuracy on AIME/HMMT and solving all IMO 2026 problems with a 7B model is striking. Ablations effectively demonstrate the necessity of the Statement Judge (showing high false certification rates without it) and the Error Judge (showing superior performance over independent resampling). The robustness check via paraphrasing helps mitigate concerns about data contamination, though the perfect scores on recent benchmarks remain a point of scrutiny.
The paper provides high reproducibility standards, including verbatim prompt templates for all components (Reasoner, Formaliser, Judges, Prover), detailed hyperparameter settings, and specific model versions (e.g., Lean v4.29.1, Mathlib). The use of open-weight models for the core pipeline (K2-Horizon, Goedel, Leanstral) further enhances reproducibility compared to closed-source baselines.
The primary limitation is the "soft certificate" nature of the output: Lean verifies the proof relative to the generated statement, but the Statement Judge is a learned proxy that can still fail, meaning the final answer is not unconditionally guaranteed correct with respect to the original natural language problem. Additionally, the computational cost is significant, requiring multiple LLM calls and Lean compilation cycles, which may limit real-time applicability. The reliance on specific, potentially proprietary or rapidly evolving model versions (like K2-Horizon) could affect long-term reproducibility.
This work bridges a critical gap in AI for Science, enabling LLMs to produce mathematically rigorous, machine-checked proofs. It sets a new standard for mathematical reasoning benchmarks and offers a template for integrating formal verification into other domains where correctness is paramount (e.g., code generation, logic synthesis). The finding that small models can solve IMO problems with verification-guided correction challenges the assumption that scale is the only driver of capability in complex reasoning tasks. Magenta introduces a training-free agentic pipeline that integrates Lean 4 formal verification into LLM mathematical reasoning, achieving 100% accuracy on AIME/HMMT benchmarks and solving all IMO 2026 problems with a 7B model by using statement adjudication and error attribution to guide self-correction.
We study how model post-training and test-time inference design affect natural-language proof generation for hard olympiad mathematics. Starting from Nemotron 3 Ultra, we train two specialist checkpoints using supervised fine-tuning and reinforcement learning, and evaluate checkpoint choice, verification, and refinement. Based on these findings, we present an open-model test-time-compute pipeline. The system operates entirely in natural language, with no formal prover, external tools, or internet access. Three Nemotron 3 Ultra checkpoints - the general-availability model and two post-trained specialists - power an iterative search that generates, verifies, and refines candidate proofs; a separate high-compute stage then selects each final submission. The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. We release the two post-trained checkpoints as well as the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.
Primary: NVIDIA
All Institutions: NVIDIA
The paper presents an open-model test-time-compute pipeline that achieves IMO Gold medal performance using natural language proofs. It contributes a rigorous methodology for iterative proof generation and verification, along with a new benchmark and open resources, marking a significant step in the capabilities of open-source LLMs for complex reasoning tasks.
The paper proposes a sophisticated test-time compute pipeline for natural language proof generation. The core methodology involves training two specialist checkpoints from Nemotron 3 Ultra using SFT and RL, then orchestrating an iterative search process that generates, verifies, and refines candidate proofs without relying on formal provers or external tools. The inclusion of a high-compute selection stage for final submissions is a key architectural decision that differentiates this from simple best-of-N sampling.
The primary result is the achievement of 30/42 points at IMO 2026, reaching the gold-medal threshold. This is a significant empirical milestone for open models. The introduction of Nemotron-IMO-Bench (200 novel problems) provides a necessary evaluation framework, though the sample size is relatively small compared to standard benchmarks. The ablation studies on checkpoint choice and verification strategies are critical to understanding the pipeline's effectiveness.
The paper explicitly states the release of post-trained checkpoints, training data, code, and submitted solutions. This high level of transparency significantly enhances reproducibility, allowing the community to verify the results and build upon the pipeline.
The reliance on a specific base model (Nemotron 3 Ultra) may limit immediate generalizability to other architectures. The benchmark size (200 problems) is small for statistical robustness. The high compute cost of the iterative search and selection stages may be prohibitive for some users.
This work demonstrates that open models can achieve human-competitive performance in high-level mathematical reasoning, potentially accelerating research in automated theorem proving and educational AI. The open release of assets fosters community development in this niche. The paper presents an open-model test-time-compute pipeline that achieves IMO Gold medal performance using natural language proofs. It contributes a rigorous methodology for iterative proof generation and verification, along with a new benchmark and open resources, marking a significant step in the capabilities of open-source LLMs for complex reasoning tasks.
Open-source large language models (LLMs) are increasingly competitive with closed-source models while offering transparency and the ability to run inference without exposing user inputs to a service provider. However, running large-scale models locally requires substantial computational resources. In practice, users may still resort to a third-party provider, giving rise to privacy and correctness concerns. Existing solutions that address these problems often impose substantial server overhead or introduce additional trust assumptions. In this paper, we present Maverick, a novel approach to private and verifiable LLM inference based on a protocol for delegating matrix-vector multiplication, a dominant operation in LLMs. At its core, Maverick provides, to our knowledge, the first information-theoretically sound verification protocol for matrix-vector multiplication delegation with transparent preprocessing, efficient (batch) verification, and virtually no server overhead. We combine this verification primitive with LPN-based pseudorandom masking to provide input privacy. We implement our matrix-vector delegation primitive and use it to build an end-to-end prototype of Maverick, which we evaluate on Qwen3-4B by measuring throughput in tokens per second. We evaluate client configurations with 1-8 threads. With one client thread and a CPU server using up to 128 threads, Maverick achieves throughput gains over local inference of up to 17x when privacy masks are generated online, 45x when they are precomputed, and 44x when only verification is required. With four client threads, the corresponding gains are 13x, 18x, and 17x. When server computation is no longer the bottleneck, client-side microbenchmarks with simulated network delay show speedups of 12x-20x, 34x-135x, and 38x-157x.
Primary: Yale University
All Institutions: Yale University
Maverick introduces a practical framework for private and verifiable LLM inference by delegating matrix-vector multiplications with minimal server overhead. The paper presents a novel information-theoretically sound verification protocol combined with LPN-based masking, achieving significant throughput gains over local inference on Qwen3-4B, thereby offering a viable path for secure, resource-efficient remote LLM usage.
The paper proposes Maverick, a system for private and verifiable LLM inference that delegates the dominant matrix-vector multiplication (MVM) operations to a server while ensuring privacy and correctness. The core contribution is a verification protocol for MVM delegation that is information-theoretically sound, features transparent preprocessing, and incurs virtually no server overhead. This is achieved by combining a novel verification primitive with LPN (Learning With Noisy) based pseudorandom masking for input privacy. The methodology is robust, addressing the two main pain points of remote inference: trust in the server's correctness and the leakage of user prompts. By shifting the heavy computational load of verification to the client (which is feasible for modern CPUs) and keeping the server's overhead minimal, the approach is architecturally sound for the specific use case of resource-constrained clients delegating to powerful servers.
The evaluation is conducted on the Qwen3-4B model, measuring throughput in tokens per second. The results are impressive, showing throughput gains of up to 17x-45x over local inference depending on the configuration (online vs. precomputed masks, verification-only vs. full privacy). The paper provides detailed breakdowns for different client thread counts (1-8 threads) and server configurations (up to 128 threads). The inclusion of microbenchmarks with simulated network delay further strengthens the evaluation by isolating client-side bottlenecks. The choice of a 4B parameter model is reasonable for a prototype, as it is large enough to be relevant but small enough to be tested on standard hardware.
The paper provides sufficient detail on the protocol design and the specific model used (Qwen3-4B). However, as an arXiv preprint with no explicit mention of a public code repository in the provided text, full reproducibility is currently limited. The authors are from Yale, a reputable institution, which increases confidence in the implementation quality, but the lack of a linked codebase is a minor drawback for immediate replication.
The primary limitation is the scope of the evaluation, which is limited to a single model size (4B parameters). It is unclear how the overhead scales for much larger models (e.g., 70B+), although the linear nature of MVM suggests it should scale well. Additionally, the reliance on LPN-based masking introduces a security parameter that must be carefully chosen to balance privacy and performance, which may not be trivial for all threat models. The paper also focuses on CPU-based servers; the applicability to GPU-accelerated servers, which are the norm for LLM inference, is not fully explored in the provided text.
This work has significant potential impact on the deployment of open-source LLMs. By enabling private and verifiable inference without requiring users to own high-end hardware, it democratizes access to powerful AI models while preserving privacy. This could be particularly impactful for mobile devices, IoT, or edge computing scenarios where local inference is too slow or resource-intensive. It also addresses the growing concern of "trust" in AI services, providing a cryptographic guarantee of correctness that is lacking in most current cloud inference APIs. Maverick introduces a practical framework for private and verifiable LLM inference by delegating matrix-vector multiplications with minimal server overhead. The paper presents a novel information-theoretically sound verification protocol combined with LPN-based masking, achieving significant throughput gains over local inference on Qwen3-4B, thereby offering a viable path for secure, resource-efficient remote LLM usage.
Driving vision-language-action (VLA) models increasingly reason before acting, but their intermediate reasoning is often weakly grounded in physical scene evidence and loosely connected to executable behavior. We present GRAVA, a framework built around Grounded Reasoning-to-Action (GRA), which unifies grounding, reasoning, and action generation in a single autoregressive stream. GRA links action-relevant language references to 2D visual regions and ego-centric physical states, organizes object interactions and decisions in a trajectory-anchored typed graph, and serializes this structure into grounded reasoning. A single VLM generates this reasoning followed by a compact Executable Planner action that is deterministically decoded into a continuous trajectory. We further introduce an agentic GRA data construction pipeline that combines forward scene grounding with backward trajectory anchoring, and use it to build GR-NavSim with 2.2M grounded question-answer pairs and 70K GRA reasoning traces. A progressive training strategy develops grounded cognition through pre-training, establishes the reasoning-to-action interface through imitation, and improves driving behavior through reinforcement learning and exploration. Using about 60% of the available human driving demonstrations for action supervision, GRAVA-8B achieves state-of-the-art performance among purely autoregressive driving models on the full NAVSIM benchmark. On an internal long-tail benchmark, full GRA improves key-object compliance and Closed-loop Driving Score by 19.3% and 20.5% over action-only prediction, respectively. These results show the benefit of preserving action-relevant physical evidence from grounded reasoning through executable action generation.
Primary: Beijing Institute of Technology
All Institutions: Beijing Institute of Technology, Shenzhen Automotive Research Institute, Shenzhen Jiguangzhijie Technology Co., Ltd., Nanyang Technological University
GRAVA introduces a unified framework for grounded reasoning-to-action in autonomous driving, achieving state-of-the-art performance on NAVSIM by integrating visual grounding, reasoning, and action generation in a single autoregressive stream with a novel trajectory-anchored graph representation and agentic data construction pipeline.
The paper proposes GRAVA, a framework that unifies visual grounding, reasoning, and action generation in a single autoregressive stream for autonomous driving. The core novelty lies in the "Grounded Reasoning-to-Action" (GRA) representation, which uses a trajectory-anchored typed graph to link linguistic references to 2D visual regions and ego-centric physical states. This structure is serialized into a reasoning sequence that directly precedes a compact "Executable Planner" action. The methodology is sound, addressing the "grounding gap" and "reasoning-to-action fragmentation" by ensuring that the physical evidence used in reasoning is explicitly connected to the final trajectory. The introduction of an agentic data construction pipeline that combines forward scene grounding with backward trajectory anchoring is a strong technical contribution, ensuring consistency between cognition and planning supervision. The progressive training strategy (pre-training, imitation, self-distillation, and Active RL) is well-structured and logically justified.
The experimental evaluation is rigorous and comprehensive. The authors benchmark GRAVA on the NAVSIM dataset, achieving state-of-the-art performance (90.48 PDMS) among purely autoregressive driving models. The ablation studies are extensive, isolating the contributions of the GRA representation, the Executable Planner, and the Active RL loop. The introduction of an internal long-tail benchmark (50K clips) to evaluate complex interactions like route obstructions and lane borrowing is a valuable addition, as public benchmarks often lack such coverage. The metrics used (PDMS, Key-Object Compliance, Closed-loop Driving Score) are appropriate for assessing both safety and progress. The results clearly demonstrate the benefit of preserving action-relevant physical evidence from grounded reasoning.
The paper provides a code repository link, which is a positive factor. However, the reliance on an "internal long-tail benchmark" limits the full reproducibility of the long-tail performance claims, as this dataset is not publicly released. The details of the agentic data construction pipeline and the specific implementation of the Active RL loop are described with sufficient detail for replication, assuming access to the nuPlan dataset and the Qwen3-VL backbone. The use of a fixed geometric decoder for the Executable Planner simplifies the action decoding process, aiding reproducibility.
The primary limitation is the dependence on the Qwen3-VL-8B backbone, which may limit the generalizability of the results to other VLM architectures. The internal long-tail benchmark, while valuable, is not publicly available, making it difficult for other researchers to verify the long-tail performance improvements. Additionally, the computational cost of the agentic data construction pipeline and the Active RL loop could be significant, potentially limiting adoption in resource-constrained settings. The paper does not extensively discuss the latency of the autoregressive reasoning process, which is a critical factor for real-time autonomous driving applications.
The paper has significant potential impact on the field of autonomous driving and vision-language-action models. By demonstrating that grounded reasoning can be effectively integrated with action generation in a single autoregressive stream, it provides a new paradigm for developing driving VLAs. The GRA representation and the agentic data construction pipeline could be adopted by other researchers to improve the grounding and reasoning capabilities of their models. The focus on long-tail scenarios and the use of reinforcement learning to refine reasoning-to-action sequences align with current trends in the field, suggesting that the work will be influential in shaping future research directions. GRAVA introduces a unified framework for grounded reasoning-to-action in autonomous driving, achieving state-of-the-art performance on NAVSIM by integrating visual grounding, reasoning, and action generation in a single autoregressive stream with a novel trajectory-anchored graph representation and agentic data construction pipeline.
Visual Autoregressive Models (VAR) generate images through next-scale prediction, producing all tokens within each scale in parallel. We show that this parallel decoding constitutes a mean-field-style approximation that discards spatial dependencies among same-scale tokens, causing locally incoherent samples regardless of backbone capacity -- a limitation of the decoding rule. Addressing this limitation, we introduce the Logit Refiner, a lightweight autoregressive module that restores intra-scale dependencies by sequentially sampling tokens conditioned on frozen backbone features. Adding only ~10% parameters and less than 5% of the base model's training compute, it plugs into any pretrained VAR checkpoint without retraining. Controlled ablations isolate joint intra-scale sampling -- rather than additional capacity or training -- as the critical ingredient. Across backbones from 310M to 2B parameters on class-conditional ImageNet 256x256, the refiner consistently improves generation quality, enabling a 1.1B-parameter model to surpass one twice its size. The approach further generalizes to text-to-image generation, confirming that the mean-field bottleneck persists across VAR variants and is effectively alleviated by our method. Project page: https://compvis.github.io/logit-refiner/
Primary: LMU Munich
All Institutions: LMU Munich, Munich Center for Machine Learning (MCML)
The paper introduces the Logit Refiner, a lightweight module that restores intra-scale spatial dependencies in Visual Autoregressive models by sequentially refining logits, significantly improving generation quality without retraining the base backbone. This work provides a clear diagnosis of the mean-field approximation limitation in parallel VAR decoding and offers an efficient, plug-and-play solution that demonstrates strong empirical gains across model scales and tasks, representing a solid contribution to the field of autoregressive image generation.
The paper identifies a specific structural weakness in Visual Autoregressive (VAR) models: the parallel decoding of tokens within a scale acts as a mean-field approximation, ignoring spatial dependencies between same-scale tokens. The proposed solution, the "Logit Refiner," is a lightweight autoregressive module that sequentially refines logits conditioned on frozen backbone features. This approach is elegant in its simplicity, requiring only ~10% additional parameters and minimal training compute, and it plugs into existing checkpoints without retraining the base model. The theoretical framing of the problem as a mean-field approximation is insightful and provides a clear motivation for the sequential refinement step.
The experiments are rigorous and well-controlled. The authors test across multiple backbone sizes (310M to 2B) on ImageNet 256x256, demonstrating consistent improvements in FID and IS metrics. A key strength is the ablation study that isolates the effect of intra-scale dependency modeling from mere increases in capacity or training data, confirming that the sequential sampling mechanism is the critical factor. The generalization to text-to-image generation further validates the robustness of the method. The claim that a 1.1B model with the refiner surpasses a 2B model without it is a strong empirical result.
The paper provides sufficient details on the architecture of the Logit Refiner and the training procedure. The project page likely contains code, which is standard for this group. The method's plug-and-play nature enhances reproducibility, as it does not require complex retraining pipelines for the base model.
The primary limitation is the increased inference latency due to the sequential refinement step, which partially negates the speed advantage of VAR's parallel decoding. While the paper notes the compute cost is low during training, the inference cost could be significant for real-time applications. Additionally, the method is specific to VAR architectures and may not directly apply to other autoregressive or diffusion-based models without modification.
This work has significant implications for the design of autoregressive generative models. By highlighting the trade-off between parallel efficiency and spatial coherence, it guides future research in hybrid decoding strategies. The plug-and-play nature of the refiner makes it an attractive tool for practitioners using existing VAR checkpoints, potentially extending the utility of current models without the need for expensive retraining. The paper introduces the Logit Refiner, a lightweight module that restores intra-scale spatial dependencies in Visual Autoregressive models by sequentially refining logits, significantly improving generation quality without retraining the base backbone. This work provides a clear diagnosis of the mean-field approximation limitation in parallel VAR decoding and offers an efficient, plug-and-play solution that demonstrates strong empirical gains across model scales and tasks, representing a solid contribution to the field of autoregressive image generation.
Scaling humanoid whole-body control toward general-purpose deployment requires large human motion corpora and training experience shared across robot bodies. Existing methods usually train one policy per robot, leaving motion experience isolated across embodiments. We introduce X-WBC, a cross-embodiment foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. Human-centered command tokens align full human motion, robot reference motion, and sparse VR observations. A causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, while lightweight robot-specific modules map the shared representation to each robot's proprioception and action space. Across nine simulated embodiments, external motions, and four real robots, experiments show that joint training improves tracking, the aligned representation supports consistent control across command sources, and the learned policy remains competitive beyond the training corpus. These results support heterogeneous humanoids as joint data sources and establish cross-embodiment joint training as a practical route toward whole-body control foundation models.
Primary: University of Washington
All Institutions: University of Washington, NVIDIA
X-WBC introduces a cross-embodiment foundation model for humanoid whole-body control that separates human motion semantics from embodiment-specific execution, enabling joint training across multiple robot bodies. The paper demonstrates that this approach improves tracking performance and generalization, providing a practical route toward general-purpose humanoid control.
The paper proposes X-WBC, a cross-embodiment foundation model for humanoid whole-body control. The core methodological contribution is the decoupling of "human motion semantics" from "embodiment-specific physical execution." It utilizes human-centered command tokens to align full human motion, robot reference motion, and sparse VR observations. A causal Transformer is employed to learn reusable temporal structures from mixed multi-robot rollouts, while lightweight robot-specific modules handle the mapping to specific proprioception and action spaces. This architecture is a logical extension of recent trends in foundation models for robotics, specifically addressing the data scarcity problem in humanoid control by leveraging large human motion datasets and sharing experience across different robot morphologies. The approach is sound and well-motivated, though the architectural components (Transformer + lightweight adapters) are not entirely new; the novelty lies in their specific application to cross-embodiment humanoid control.
The experimental setup is robust, covering nine simulated embodiments, external motions, and four real robots. The results demonstrate that joint training improves tracking performance compared to single-embodiment baselines. The paper shows that the aligned representation supports consistent control across different command sources (human motion, VR, etc.) and that the policy generalizes beyond the training corpus. The inclusion of real-world experiments on four different robots is a strong point, validating the practicality of the cross-embodiment approach. The evaluation metrics likely include tracking error, success rate, and robustness to perturbations, which are standard and appropriate for this domain.
The paper provides sufficient detail on the architecture and training procedure. However, the specific hyperparameters, dataset sizes, and code availability are not explicitly detailed in the provided text snippet. Assuming standard practices for CoRL papers, code is likely released, but without explicit confirmation in the text, reproducibility is rated as moderate-to-high. The use of standard simulation environments (likely MuJoCo or Isaac Gym) and common datasets (like AMASS or CMU) aids reproducibility.
The primary limitation is the reliance on simulation for the majority of the training data, which may introduce sim-to-real gaps that are not fully addressed. The "lightweight robot-specific modules" might require significant tuning for new embodiments, potentially limiting the "foundation" aspect. Additionally, the paper focuses on humanoid robots, so the applicability to other robot types (e.g., quadrupeds, manipulators) is not explored. The computational cost of training a foundation model across multiple embodiments could be high.
This work contributes to the growing field of foundation models for robotics, aiming to create general-purpose controllers that can be adapted to various robot bodies. It has the potential to accelerate the deployment of humanoid robots by reducing the need for extensive robot-specific data collection and training. The cross-embodiment approach could be extended to other domains, such as multi-agent control or heterogeneous robot swarms. X-WBC introduces a cross-embodiment foundation model for humanoid whole-body control that separates human motion semantics from embodiment-specific execution, enabling joint training across multiple robot bodies. The paper demonstrates that this approach improves tracking performance and generalization, providing a practical route toward general-purpose humanoid control.
Generalizable bimanual robotic manipulation requires a reusable task prior that can persist across increasingly diverse tasks, objects, scenes, embodiments, and execution conditions, thus avoiding the prohibitive cost of large-scale teleoperated demonstrations and policy retraining. In this work, we present VLBiMan++, an extended framework that expands the generalization boundary of vision-language anchored one-shot bimanual manipulation. Starting from a single human demonstration, VLBiMan++ performs task-aware decomposition to identify reusable and adaptable skill components, and employs vision-language grounded geometric adaptation to transfer these skills to novel configurations without retraining. Building on this foundation, we systematically extend generalization along five dimensions: task generalization through diverse and long-horizon skill compositions; object generalization across unseen categories, varying geometries, and more complex articulated or deformable objects; scene generalization under clutter, occlusion, and dynamic interference; embodiment generalization across heterogeneous dual-arm robotic platforms; and deployment generalization through prolonged closed-loop execution under repeated external perturbations. To support this broader scope, we further introduce object-state-aware adaptation and lightweight trajectory optimization mechanisms that accommodate changes beyond simple rigid 6-DoF pose variations while preserving reliable bimanual coordination. Extensive real-world experiments demonstrate that VLBiMan++ maintains strong task success and adaptation capability across these increasingly challenging settings. Overall, VLBiMan++ advances one-shot bimanual manipulation from demonstrating isolated transferability toward a more systematic and scalable framework for generalization across tasks, objects, scenes, embodiments, and long-term deployment conditions.
Primary: Shenzhen University
All Institutions: Shenzhen University, The Chinese University of Hong Kong, Shenzhen, DexForce, Imperial College London
The paper presents VLBiMan++, a framework that enhances one-shot bimanual manipulation by using vision-language grounding to decompose and adapt skills across diverse tasks, objects, and embodiments. While the systematic approach to generalization across five dimensions is commendable and the inclusion of deformable/articulated objects is notable, the lack of detailed quantitative results in the provided text and the reliance on human-in-the-loop decomposition limit its technical impact and reproducibility, placing it as a solid but incremental contribution to the field of robotic manipulation.
The paper proposes VLBiMan++, a framework for one-shot bimanual manipulation that leverages vision-language models (VLMs) to ground task semantics and decompose a single human demonstration into reusable skill components. The core methodological contribution is a "task-aware decomposition" that separates state-invariant skills from state-conditioned ones, followed by "vision-language grounded geometric adaptation" to transfer these skills to novel configurations without retraining. The approach extends generalization across five dimensions: task, object, scene, embodiment, and deployment. While the use of VLMs for grounding is increasingly common, the specific mechanism for decomposing a single trajectory into adaptable geometric skills and the "object-state-aware adaptation" for non-rigid objects represent a solid incremental advance in the one-shot learning paradigm for robotics. The methodology relies heavily on the quality of the initial decomposition and the accuracy of the VLM's grounding, which are critical bottlenecks.
The paper claims "extensive real-world experiments" demonstrating generalization across diverse tasks, objects (rigid, articulated, deformable), scenes (cluttered, dynamic), and embodiments. However, the provided text lacks specific quantitative results, baselines, or statistical significance tests. Without concrete numbers (e.g., success rates compared to baselines like VLBiMan or other imitation learning methods), it is difficult to rigorously assess the magnitude of improvement. The scope of experiments is broad, which is a strength, but the lack of detailed data in the abstract/conclusion summary makes the evaluation feel somewhat anecdotal. The inclusion of deformable and articulated objects is a strong point, as these are notoriously difficult for one-shot methods.
The paper provides a project link, which is a positive step for reproducibility. However, the specific details of the "lightweight trajectory optimization" and the exact VLM architecture used for grounding are not detailed in the provided text. The reliance on "human refinement" for task decomposition (mentioned in limitations) suggests that the system is not fully autonomous and may be difficult to reproduce without the specific expertise of the authors to perform the manual segmentation steps. The hardware dependencies (specific dual-arm platforms) also limit immediate reproducibility for labs without similar setups.
The authors honestly list several limitations: 1) The object-state representation is compact and may fail on highly complex/deformable topologies. 2) The closed-loop mechanism lacks comprehensive failure-aware reasoning and recovery. 3) Collision handling is limited to pre-grasp compensation. 4) Task decomposition requires human refinement, limiting scalability. 5) Hardware constraints (fixed-base, parallel grippers, no force sensing) limit dexterity. These are significant weaknesses that prevent the system from being a general-purpose solution.
The work contributes to the growing field of vision-language-action (VLA) models and one-shot imitation learning. By focusing on bimanual manipulation and generalization across embodiments, it addresses a critical gap in robotic autonomy. The framework could be adopted by researchers looking to reduce data collection costs in manipulation tasks. However, the need for human-in-the-loop decomposition and the specific hardware constraints limit its immediate widespread adoption in industry or diverse academic labs. The paper presents VLBiMan++, a framework that enhances one-shot bimanual manipulation by using vision-language grounding to decompose and adapt skills across diverse tasks, objects, and embodiments. While the systematic approach to generalization across five dimensions is commendable and the inclusion of deformable/articulated objects is notable, the lack of detailed quantitative results in the provided text and the reliance on human-in-the-loop decomposition limit its technical impact and reproducibility, placing it as a solid but incremental contribution to the field of robotic manipulation.
Integrating contact information into visuomotor policies remains an open problem. Touch is essential to robust manipulation, yet most modern policies, including pretrained vision-language-action (VLA) models, operate from vision and proprioception alone. Existing approaches to closing this gap require specialized tactile hardware, add separate tactile encoders, or commit to non-image policy backbones, all incompatible with the modern paradigm of image-conditioned policies built on pretrained 2D visual representations. Our key insight is that the bottleneck is not the contact information itself, but how it is delivered: when contact signals are exposed in the same spatial frame as the scene the policy already attends to, they become directly usable by any image-conditioned policy without architectural changes. We operationalize this insight in Visible Touch, paired with a custom low-cost magnetic contact sensor that is open-sourced and fabricated from off-the-shelf parts via a parametric CAD-to-mold pipeline. Across the LIBERO benchmark, Visible Touch improves BC-Transformer success by 15.7 percentage points on average in the 2-view setting, with similar gains in the 1-view setting; controlled comparisons show that the contact-integration strategy strongly affects how effectively tactile information is used. The pattern holds when fine-tuning pretrained VLAs: miniVLA on LIBERO gains 25 percentage points on average, and $π_{0.5}$ on four real-world contact-rich tasks gains 30 percentage points with our custom sensor.
Primary: University of California, Los Angeles
All Institutions: University of California, Los Angeles
The paper introduces a visual overlay method for integrating tactile contact information into image-conditioned visuomotor policies, demonstrating significant performance gains on both simulation and real-world manipulation benchmarks. This approach effectively decouples the tactile sensing hardware from the policy architecture, allowing standard vision-based models to leverage contact data without modification, which is a practical and impactful contribution to the field of robotic manipulation.
The paper proposes "Visible Touch," a method that renders contact information as visual overlays in the same spatial frame as the scene. This approach allows standard image-conditioned policies (like BC-Transformer and VLA models) to utilize tactile data without architectural modifications. The method is paired with a custom, low-cost magnetic contact sensor that is open-sourced and fabricated via a parametric CAD-to-mold pipeline. The core insight is that the bottleneck for integrating touch is not the information content but the delivery mechanism; aligning the signal with the visual attention space makes it directly usable by existing vision-based backbones.
The authors evaluate the method on the LIBERO benchmark, reporting a 15.7 percentage point improvement for BC-Transformer in the 2-view setting. They also demonstrate gains when fine-tuning pretrained VLAs, specifically miniVLA (25 pp gain on LIBERO) and $\pi_{0.5}$ (30 pp gain on four real-world contact-rich tasks). Controlled comparisons highlight that the contact-integration strategy significantly impacts the effectiveness of tactile information usage. The results are strong, showing consistent improvements across both simulation and real-world settings.
The paper claims high reproducibility by open-sourcing the custom magnetic contact sensor and providing a parametric CAD-to-mold pipeline for fabrication from off-the-shelf parts. The project website is provided. However, the full paper text provided in the prompt is largely a stub with section headers, so specific implementation details of the rendering algorithm and policy training are not visible in the provided text, though the abstract implies a clear methodology.
The provided text is incomplete, containing only the abstract and section headers, which limits a deep critique of the specific rendering algorithms or failure cases. The reliance on a custom sensor, even if low-cost, may limit immediate adoption by groups without fabrication capabilities. The generalization to other types of tactile sensors or non-magnetic contact scenarios is not discussed in the available text.
This work bridges the gap between tactile sensing and the dominant paradigm of vision-language-action models. By providing a drop-in solution for integrating touch into existing image-based policies, it lowers the barrier to entry for robust manipulation research. The open-sourcing of the hardware is a significant contribution to the community, enabling broader experimentation with contact-rich manipulation tasks. The paper introduces a visual overlay method for integrating tactile contact information into image-conditioned visuomotor policies, demonstrating significant performance gains on both simulation and real-world manipulation benchmarks. This approach effectively decouples the tactile sensing hardware from the policy architecture, allowing standard vision-based models to leverage contact data without modification, which is a practical and impactful contribution to the field of robotic manipulation.
Hamilton-Jacobi (HJ) reachability provides a principled framework for synthesizing safety certificates and robust controllers for safety-critical robotic systems. However, applying reachability analysis to high-dimensional nonlinear systems remains challenging: classical grid-based solvers suffer from the curse of dimensionality, continuous-time neural solvers require accurate spatial value gradients, and reinforcement-learning-based approaches often suffer from weak boundary anchoring and non-stationary adversarial policy optimization. We propose a discrete-time neural reachability framework for control-disturbance-affine systems that learns backward reachable tubes (BRTs) and backward reach-avoid tubes (BRATs) through Bellman-Isaacs value propagation. Our key idea is to combine equation-driven self-supervision with structured policy learning: rather than computing explicit PDE-gradients, we exploit the bang-bang structure of optimal safety interventions to construct approximate teacher actions from gradient-free value probes, converting adversarial actor learning into supervised policy learning. To stabilize long-horizon value propagation, we leverage the learned actor to train the value function backward from the terminal boundary using a windowed temporal curriculum, where each window is used as the boundary condition for the next window. Across benchmark problems up to 80 dimensions, our method learns accurate reachability value functions while improving stability over existing learning-based solvers. We further demonstrate observation-space scalability on F1-tenth racing with over 16,000-dimensional egocentric inputs. The learned safety filter generalizes zero-shot to unseen tracks and transfers to a physical RC car, achieving real-time robust collision avoidance.
Primary: University of Michigan
All Institutions: University of Michigan
The paper introduces a gradient-free neural Hamilton-Jacobi reachability framework that leverages bang-bang structure for supervised policy learning and windowed temporal curricula for stability. This approach effectively scales reachability analysis to high-dimensional systems (up to 80 dimensions) and demonstrates practical utility through zero-shot generalization and physical robot deployment, representing a significant step forward in scalable safety-critical control.
The paper proposes a discrete-time neural framework for Hamilton-Jacobi (HJ) reachability analysis, specifically targeting control-disturbance-affine systems. The core innovation is the avoidance of explicit PDE gradient computation, which is typically required for continuous-time neural solvers and is numerically unstable. Instead, the authors exploit the bang-bang structure of optimal safety interventions to construct "teacher actions" via gradient-free value probes. This converts the traditionally non-stationary adversarial actor learning problem into a supervised policy learning task. Additionally, a windowed temporal curriculum is introduced to stabilize long-horizon value propagation by using the learned actor to train the value function backward from the terminal boundary. This approach effectively addresses the curse of dimensionality and the instability issues associated with standard grid-based or continuous-time neural solvers.
The experimental evaluation is strong, covering benchmark problems up to 80 dimensions, which is significant for reachability analysis. The paper demonstrates scalability in observation space using the F1-tenth racing simulator with over 16,000-dimensional egocentric inputs. Crucially, the work extends beyond simulation to a physical RC car, demonstrating real-time robust collision avoidance and zero-shot generalization to unseen tracks. This combination of high-dimensional simulation and physical hardware validation provides strong evidence for the practical utility of the method.
The paper provides a link to the code repository (https://github.com/TheZeyuanFeng/grad_free_reach/tree/main) and a project webpage. The availability of code is a positive factor for reproducibility. However, without access to the full implementation details in the text provided (which appears to be a summary/abstract with section headers but missing the detailed body text), a full assessment of hyperparameters and specific implementation nuances is limited. The presence of code is the primary indicator here.
The method is restricted to control-disturbance-affine systems, which may not cover all complex robotic dynamics. The reliance on the bang-bang structure for teacher action construction assumes specific properties of the optimal safety intervention that may not hold in all nonlinear scenarios. Furthermore, the "windowed temporal curriculum" introduces a sequential training dependency that could limit parallelization opportunities during the learning phase.
This work has significant implications for the safety of autonomous systems, particularly in high-dimensional state spaces where traditional reachability analysis is intractable. By enabling scalable, gradient-free neural reachability, it opens the door to applying rigorous safety certificates to complex robots like autonomous vehicles and humanoid robots. The transfer to physical hardware suggests immediate applicability in real-world safety-critical applications. The paper introduces a gradient-free neural Hamilton-Jacobi reachability framework that leverages bang-bang structure for supervised policy learning and windowed temporal curricula for stability. This approach effectively scales reachability analysis to high-dimensional systems (up to 80 dimensions) and demonstrates practical utility through zero-shot generalization and physical robot deployment, representing a significant step forward in scalable safety-critical control.
Post-training vision-language-action (VLA) models for specific robots and tasks requires in-domain demonstrations, yet collecting diverse robot data is costly. Egocentric human demonstrations provide a scalable alternative, but directly mixing human and robot data can introduce cross-embodiment discrepancies and degrade policy performance. To address this challenge, we introduce ReWeight, a framework that incorporates human data into VLA post-training through demonstration-level retrieval and sample-level weighting. ReWeight learns a cross-embodiment visuomotor representation that combines visual observations with future actions to measure behavioral similarity between human and robot demonstrations. Based on optimal transport, it retrieves human demonstrations relevant to the target robot data and assigns larger weights to samples with smaller cross-embodiment discrepancies. We evaluate ReWeight using $π_{0.5}$ across eight simulation tasks and four real-world tasks under both clean and randomized settings. In simulation, ReWeight improves the average success rate of post-trained $π_{0.5}$ from 39% with only robot data and 44% with randomly mixed human-robot data to 57%. In the physical experimental setting, it achieves an average success rate of 68.8%, outperforming the baselines by 28.8% and 13.8%, respectively. Overall, ReWeight provides an effective paradigm for transforming abundant egocentric human experience into transferable supervision for robot learning. (Project webpage: https://reweight-vla.github.io/)
Primary: The University of Hong Kong
All Institutions: The University of Hong Kong, Institute of Artificial Intelligence (TeleAI) China Telecom, Shenzhen Research Institute of Northwestern Polytechnical University
ReWeight introduces a retrieval and weighting mechanism to effectively leverage human demonstrations for VLA post-training. The paper presents a solid methodological contribution with strong empirical results, demonstrating significant improvements in robot manipulation tasks by mitigating cross-embodiment discrepancies through optimal transport-based sample weighting.
The paper proposes ReWeight, a framework for post-training Vision-Language-Action (VLA) models by integrating egocentric human data. The core method involves learning a cross-embodiment visuomotor representation that fuses visual observations with future actions to quantify behavioral similarity. It employs optimal transport to retrieve relevant human demonstrations and applies sample-level weighting to mitigate cross-embodiment discrepancies. The approach is logically sound, addressing a known bottleneck in robot learning (data scarcity) by leveraging scalable human data.
Experiments are conducted on $\pi_{0.5}$ across 8 simulation tasks and 4 real-world tasks. The results show a significant improvement in success rates (from 39% to 57% in simulation, and 68.8% in real-world settings) compared to baselines using only robot data or randomly mixed data. The evaluation covers both clean and randomized settings, providing a reasonably robust assessment of the method's effectiveness.
The paper provides a project webpage, which likely contains code or further details. However, as an arXiv preprint, full reproducibility depends on the availability of the specific VLA model checkpoints and the human/robot datasets used, which may not be fully open. The description of the optimal transport mechanism is high-level, and specific hyperparameters for the weighting scheme are not detailed in the provided text.
The method relies on the quality of the learned cross-embodiment representation; if the visual or action spaces are too disparate, the similarity metric may fail. The evaluation is limited to a specific VLA model ($\pi_{0.5}$) and a small number of tasks, limiting the generalizability of the findings. The computational cost of optimal transport for large-scale datasets is not discussed.
The work contributes to the growing field of cross-embodiment learning, offering a practical pathway to utilize abundant human video data for robot policy improvement. It could influence future research on data-efficient robot learning and the integration of human-robot collaboration data. ReWeight introduces a retrieval and weighting mechanism to effectively leverage human demonstrations for VLA post-training. The paper presents a solid methodological contribution with strong empirical results, demonstrating significant improvements in robot manipulation tasks by mitigating cross-embodiment discrepancies through optimal transport-based sample weighting.
Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a representation of action-dependent scene changes. We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual observations, goals, and actions as discrete tokens. By varying conditioning and target spans, the same model learns action prediction, action-conditioned next-observation prediction, terminal goal-state prediction, and trajectory-to-instruction reconstruction. These interfaces support test-time scaling through goal prediction, action-candidate evaluation, and joint refinement of action and future-state predictions. We continually pretrain the model on approximately 1.33 million trajectories from 48 Open X-Embodiment datasets and adapt it separately to downstream domains. On two VLABench tasks, robot pretraining improves adaptation within a fixed Stage-2 step budget, and the full objective mixture improves shifted-instruction success over Policy-only post-training under the same coupled decoder. Combining goal guidance with joint action-next-state denoising further improves shifted-instruction success over action-only decoding; the benefit depends on how the predictions are composed. Dynin-Robotics achieves competitive performance on LIBERO and zero-shot LIBERO-Plus, together with a 78.4% average success rate across four manipulation conditions on a Franka Research 3 robot. An optimized block-parallel implementation accelerates model-side action decoding by up to 29.2x relative to the base implementation under the reported profiling setup. These results support shared trajectory modeling as a common interface for learning complementary robot objectives and composing their predictions during control.
Primary: Unknown
All Institutions: Unknown
Dynin-Robotics introduces a unified omnimodal diffusion model for robot control that jointly learns action, goal, and next-state prediction, achieving competitive performance on standard benchmarks and physical robots. The paper presents a solid technical contribution with a unified training objective and practical implementation optimizations, though its impact is currently limited by the lack of peer-reviewed venue and community adoption.
The paper proposes Dynin-Robotics, a unified diffusion-based Vision-Language-Action (VLA) model that treats language, visual observations, goals, and actions as discrete tokens within a shared trajectory model. The core methodological contribution is the use of an "omnimodal masked-diffusion backbone" (Dynin-Omni) to jointly learn action prediction, action-conditioned next-observation prediction, terminal goal-state prediction, and trajectory-to-instruction reconstruction. This unified formulation allows for test-time scaling through mechanisms like goal prediction and joint refinement of action and future-state predictions. The approach is technically sound, leveraging the strengths of diffusion models for multimodal generation and the discrete tokenization for efficient processing. The inclusion of a block-parallel implementation for accelerated decoding is a practical engineering contribution that addresses a common bottleneck in diffusion-based robotics models.
The model is pre-trained on a substantial dataset of approximately 1.33 million trajectories from 48 Open X-Embodiment datasets. Evaluation is conducted on VLABench, LIBERO, zero-shot LIBERO-Plus, and a physical Franka Research 3 robot. The results show competitive performance, with a 78.4% average success rate on the physical robot across four manipulation conditions. The paper demonstrates that the full objective mixture improves shifted-instruction success compared to policy-only post-training, and that combining goal guidance with joint denoising further improves performance. The 29.2x speedup in action decoding is a significant practical improvement. However, the citation count is 0, and the venue is arXiv, suggesting the work is recent or pre-publication, which limits the assessment of its long-term impact.
The paper provides details on the training recipe, dataset sources (Open X-Embodiment), and implementation optimizations (block-parallel). However, without access to the code repository (not provided in the text) or specific hyperparameter settings, full reproducibility is difficult to assess. The use of standard benchmarks (LIBERO, VLABench) aids in comparability with other works.
The primary limitation is the lack of clear institutional affiliation and venue information, which makes it difficult to gauge the rigor of peer review. The citation count of 0 indicates the paper has not yet gained traction in the community. The reliance on discrete tokenization for continuous actions and observations may introduce quantization errors, although the paper does not explicitly discuss this trade-off in detail. The performance gains, while positive, are described as "competitive" rather than state-of-the-art, suggesting there is still room for improvement.
The work contributes to the growing field of unified VLA models, which aim to bridge the gap between perception, language understanding, and action generation. The proposed test-time scaling techniques could be applicable to other robotic tasks and potentially other domains requiring multimodal sequence generation. The acceleration of diffusion model decoding is a valuable contribution for real-time robotic control. Dynin-Robotics introduces a unified omnimodal diffusion model for robot control that jointly learns action, goal, and next-state prediction, achieving competitive performance on standard benchmarks and physical robots. The paper presents a solid technical contribution with a unified training objective and practical implementation optimizations, though its impact is currently limited by the lack of peer-reviewed venue and community adoption.
Humanoid locomotion on granular terrain remains a significant challenge due to its complex foot-terrain interaction dynamics that are difficult to model. Existing approaches either ignore granular contact dynamics or incorporate simplified normal force models with heuristic tangential components. In this work, we present a physics-grounded granular contact model based on three-dimensional resistive force theory (3D RFT) and efficiently simulate granular terrain for reinforcement learning (RL) training. Unlike traditional rigid contact models and simplified granular contact models with ad-hoc heuristics, our contact solver produces physically accurate granular intrusion dynamics without resorting to heuristics. It captures realistic penetration and tangential drag during training, enabling the policy to learn behaviors that transfer reliably to real-world granular terrain where rigid contact models fail. To adapt to varying terrain conditions, we train a terrain-adaptive locomotion controller via teacher-student RL, using a variational autoencoder to encode terrain information into a compact latent representation. Simulation studies using material point method (MPM) with NVIDIA Newton demonstrate that our method generalizes to unseen granular terrains, achieves a significantly higher success rate than baselines, and demonstrates zero-shot terrain identification and adaptation. We further validate our approach through extensive hardware experiments across diverse real-world granular terrains including basalt, dry sand, and beach sand. To the best of our knowledge, this is the first demonstration of agile humanoid locomotion on real-world granular terrain. Project page: https://humanoid-gm-locomotion.github.io/HUMANOID-GM/
Primary: Georgia Institute of Technology
All Institutions: Georgia Institute of Technology, Northeastern University
The paper introduces a physics-grounded granular contact model using 3D RFT and a teacher-student RL framework to enable agile humanoid locomotion on real-world granular terrain. This is a significant contribution to robotics and reinforcement learning, as it addresses the critical sim-to-real gap in complex, deformable terrain interactions, demonstrating superior performance and generalization compared to heuristic-based approaches.
The paper proposes a physics-grounded granular contact model based on 3D Resistive Force Theory (RFT) to simulate granular terrain for reinforcement learning (RL). This is a significant methodological improvement over heuristic or rigid-body approximations, as it captures realistic penetration and tangential drag. The authors employ a teacher-student RL framework where a Variational Autoencoder (VAE) encodes terrain properties into a compact latent representation, allowing the student policy to adapt to varying terrain conditions without explicit terrain labels. The use of the Material Point Method (MPM) with NVIDIA Newton for high-fidelity simulation is a strong technical choice that bridges the gap between physical accuracy and computational efficiency for RL training.
The evaluation is robust, featuring both simulation studies and extensive hardware experiments. The simulation results demonstrate generalization to unseen granular terrains and zero-shot terrain identification. Crucially, the paper validates its approach on real-world hardware across diverse terrains (basalt, dry sand, beach sand), claiming to be the first demonstration of agile humanoid locomotion on real-world granular terrain. The comparison against baselines shows significantly higher success rates, providing strong empirical evidence for the efficacy of the physics-grounded contact model.
The paper references specific tools (NVIDIA Newton, MPM) and provides a project page. However, the full text provided is a draft with placeholder sections (e.g., "First_draft/Method"), which limits the ability to verify specific hyperparameters or implementation details. The reliance on proprietary simulation tools (NVIDIA Newton) may pose some barriers to reproducibility for labs without access to such resources, though the physics-based approach (RFT) is standard.
The primary limitation is the reliance on a specific, potentially proprietary simulation stack (NVIDIA Newton). Additionally, the "first demonstration" claim, while impressive, is limited to a specific set of granular terrains; the generalization to other granular materials (e.g., snow, mud) is not explicitly tested in the provided text. The computational cost of MPM simulations, even with optimizations, remains high compared to rigid-body dynamics, which may limit real-time deployment on less powerful hardware.
This work has high potential impact in the field of legged robotics and humanoid control. By providing a physically accurate method for simulating granular terrain, it enables the development of more robust locomotion policies that can operate in unstructured environments. The teacher-student approach with latent terrain encoding offers a scalable path for adapting to diverse terrains, which is a critical challenge for deploying humanoid robots in the real world. The paper introduces a physics-grounded granular contact model using 3D RFT and a teacher-student RL framework to enable agile humanoid locomotion on real-world granular terrain. This is a significant contribution to robotics and reinforcement learning, as it addresses the critical sim-to-real gap in complex, deformable terrain interactions, demonstrating superior performance and generalization compared to heuristic-based approaches.
A hallmark of human dexterity is the cooperative use of fingers, where different fingers take on distinct yet coordinated roles to accomplish fine manipu- lation, such as capping a pen with the hand that holds it. We study this finger-level coordination through in-hand assembly: mating two rigid objects within a single dexterous hand, with no second arm and no fixture. We present a reinforcement learning formulation to solve this problem in a unified framework, which is driven by a goal relative pose between the two parts. Finger coordination is shaped by a function-based auxiliary reward and regularized toward a single human reference pose, while domain randomization and a fusion of historical proprioception and object observation confer robustness to occlusion-induced estimation noise. The same recipe solves three different assembly tasks (Bottle, Syringe, and Marker). Trained purely in simulation, the policies transfer zero-shot to hardware with a single camera, demonstrating robustness to state-estimation errors caused by oc- clusion. Our experiments also reveal that in-hand assembly places demands on hand morphology and can serve as a benchmark for modern robotic hand systems. Videos and code are available at https://ltbgbird.github.io/in-hand-assembly-page/.
Primary: The University of Hong Kong
All Institutions: The University of Hong Kong, The Hong Kong University of Science and Technology (Guangzhou), The Hong Kong University of Science and Technology, ETH Zurich
The paper introduces a robust reinforcement learning framework for fixture-free in-hand assembly, demonstrating that a single dexterous hand can coordinate multiple fingers to mate two objects with zero-shot sim-to-real transfer. By combining function-based reward shaping with human pose priors and robust state estimation, the work establishes a new benchmark for dexterous manipulation and provides critical insights into the morphological requirements for advanced robotic hands.
The paper proposes a reinforcement learning framework for "in-hand assembly," a task where a single dexterous hand manipulates two objects to mate them without external fixtures. The core methodological contribution is a unified RL formulation driven by relative pose goals, augmented with a function-based auxiliary reward to shape finger coordination (distinguishing roles like pinch vs. grasp) and a regularization term toward human reference poses. The approach relies on sim-to-real transfer using domain randomization and a fusion of proprioception and visual object state estimation. While the use of RL for dexterous manipulation is established, the specific formulation for multi-object coordination within a single hand, particularly the reward shaping for finger specialization, is a meaningful technical advance. The handling of occlusion-induced noise via historical proprioception fusion is also a solid engineering solution to a critical real-world problem.
The experiments are rigorous and well-designed. The authors evaluate three distinct tasks (Bottle, Syringe, Marker) to demonstrate generality. A key strength is the "Hand Morphology Benchmark" section, which tests the pipeline on four different robotic hands (Sharpa, Wuji, Allegro, XHand), providing valuable insights into how hardware constraints affect dexterity. The ablation studies clearly isolate the contributions of vision, recurrent architecture (LSTM vs. MLP), and specific reward components. Real-world validation is strong, with zero-shot transfer from simulation to hardware using only a single camera. The comparison between closed-loop control and open-loop replay convincingly demonstrates the necessity of feedback. The robustness tests against perturbation and wrist tilt further solidify the practical utility of the learned policies.
The paper provides high reproducibility. It details the simulation setup (IsaacSim, PhysX parameters), policy architecture (LSTM/MLP dimensions), RL hyperparameters (PPO settings), and domain randomization ranges. The code and videos are available on the project website. The description of the real-world deployment, including camera calibration and pose tracking stabilization, is detailed enough for replication. The use of standard hardware (Sharpa Wave, Franka, RealSense) and open-source simulation tools lowers the barrier to entry for other researchers.
The primary limitation is the manual initialization of objects; the system requires a human to place the objects in the hand before control is activated, meaning it does not solve the full pipeline of grasping from a table. The paper also notes that current rigid-body simulators struggle with planar contacts, leading to sim-to-real gaps in pinch stability at certain wrist tilts. The scope is limited to three specific assembly tasks, and the generalization to arbitrary part geometries (e.g., threading, friction-fit) is left for future work.
This work significantly advances the field of dexterous manipulation by demonstrating that a single hand can perform complex, coordinated multi-object tasks. It provides a valuable benchmark for evaluating robotic hand morphology and control algorithms. The insights into finger specialization and the necessity of closed-loop feedback under occlusion will influence the design of future dexterous systems. The zero-shot sim-to-real transfer capability is particularly impactful for reducing the cost and time of real-world deployment. The paper introduces a robust reinforcement learning framework for fixture-free in-hand assembly, demonstrating that a single dexterous hand can coordinate multiple fingers to mate two objects with zero-shot sim-to-real transfer. By combining function-based reward shaping with human pose priors and robust state estimation, the work establishes a new benchmark for dexterous manipulation and provides critical insights into the morphological requirements for advanced robotic hands.
Robotic manipulation tasks are often decomposed into behaviors or skills. However, one often needs to predefine these behaviors for specific tasks or try to cover a wide range of tasks using generic skills. As a result, these behaviors can remain too coarse to expose the geometric, control, and scene-dependent decisions required for execution. We introduce Grounded Task Axes v2 (GTA-2), a modular multi-VLM framework that constructs executable, task-bespoke manipulation skills from reusable object-centric task-axis components. Rather than predicting actions end-to-end or composing fixed task-level primitives, GTA-2 represents each skill as semantic subtasks comprising task-relevant keypoints and axes, controller compositions, and scene-dependent parameters. Four specialized VLM agents separately decompose the task, construct an abstract task-axis skill, assign controller parameters, and ground the required visual features from RGB-D observations. This abstraction-to-grounding factorization enables zero-shot skill generation without task-specific robot demonstrations, policy training, or fine-tuning. It also keeps intermediate decisions explicit, allowing targeted human feedback to refine an incorrect stage while preserving correct components. We evaluate GTA-2 on 14 real-robot manipulation tasks against a VLA policy pi_{0.5} and two Code-as-Policies baselines using task-axis controllers or conventional robot primitives. GTA-2 achieves an average zero-shot success rate of 73.9%, exceeding the strongest baseline by 31.4 percentage points, while targeted refinement raises GTA-2's average success rate to 90.7%. Project page: https://gta2-project.github.io/
Primary: Carnegie Mellon University
All Institutions: Carnegie Mellon University, Bosch Research
GTA-2 introduces a modular multi-VLM framework that synthesizes executable robot manipulation skills via grounded task axes, achieving high zero-shot success rates on real-robot tasks by factorizing skill generation into explicit, refinable stages. The paper presents a compelling alternative to end-to-end VLA policies by offering a transparent, component-based approach to skill synthesis, with strong empirical results on a real-robot benchmark, though the limited task diversity and reliance on human refinement for optimal performance are notable constraints.
The paper proposes GTA-2, a modular multi-VLM framework that decomposes robotic manipulation into "task axes" (keypoints, axes, controller compositions). The core innovation is the factorization of skill generation into four specialized VLM agents: task decomposition, abstract skill construction, parameter assignment, and visual grounding. This approach avoids end-to-end action prediction, instead generating explicit, inspectable intermediate representations that map to low-level controllers. The methodology is sound in its modular design, allowing for targeted human-in-the-loop refinement of specific stages (e.g., correcting a wrong keypoint without re-running the entire pipeline). The use of "task axes" as a semantic bridge between high-level language and low-level control is a creative contribution to the field of robot skill synthesis.
The evaluation is conducted on 14 real-robot manipulation tasks, which is a significant practical benchmark. The comparison against a VLA policy (pi_0.5) and Code-as-Policies baselines is relevant. The reported zero-shot success rate of 73.9% (outperforming the strongest baseline by 31.4 points) and the post-refinement rate of 90.7% are strong empirical results. However, the sample size of 14 tasks is relatively small for a general claim of "zero-shot skill generation," and the specific nature of the tasks (likely constrained to manipulation) limits the generalizability of the results to broader robotic domains.
The paper provides a project page, which likely contains code or further details. The modular nature of the VLM agents suggests that the pipeline is reproducible if the specific VLM models and controller libraries are available. However, the "task-axis" representation is novel, and without extensive documentation on how these axes are defined and mapped to controllers, full reproduction by external groups may be challenging. The reliance on specific VLM capabilities for grounding also introduces variability depending on the underlying model versions.
The primary limitation is the scope of the evaluation (14 tasks). While the success rates are high, it is unclear how the system performs on highly dynamic or unstructured environments beyond the tested set. The "zero-shot" claim relies on the VLMs' pre-trained knowledge, which may fail on novel object categories or task types not seen during VLM pre-training. Additionally, the human-in-the-loop refinement, while effective, introduces a dependency on human expertise for error correction, which may not scale to fully autonomous deployment.
This work contributes to the trend of using VLMs for high-level planning in robotics. By providing an explicit, editable intermediate representation (task axes), it offers a path toward more transparent and debuggable robotic systems compared to black-box VLA policies. This could be impactful for industrial applications where explainability and safety are critical. The collaboration between CMU and Bosch Research suggests potential for real-world industrial adoption. GTA-2 introduces a modular multi-VLM framework that synthesizes executable robot manipulation skills via grounded task axes, achieving high zero-shot success rates on real-robot tasks by factorizing skill generation into explicit, refinable stages. The paper presents a compelling alternative to end-to-end VLA policies by offering a transparent, component-based approach to skill synthesis, with strong empirical results on a real-robot benchmark, though the limited task diversity and reliance on human refinement for optimal performance are notable constraints.
Determinism and numerical reproducibility are increasingly required of GPU kernels in machine learning systems, yet deterministic implementations of the same kernel can still differ bit for bit. Floating-point reduction order is the primary cause, alongside partial-sum precision, fused multiply-add operations, and rounding placement. These choices may be hand-coded, selected by a block-level language such as Triton, or hidden inside a closed-source library such as cuBLAS or rocBLAS. A tile shape chosen for performance therefore also determines the arithmetic, potentially breaking batch invariance. Preserving a fixed order can cost up to 20 percent, while an autotuner cannot identify which configurations are bitwise equivalent. We characterize the factors determining the bitwise behavior of reductions and general matrix multiplication (GEMM). First, we introduce a descriptor of GEMM reduction order, including the partitioning of K in split-K GEMM. Using it, we perform the first black-box reconstruction of a closed-source library's arithmetic for bit-level correctness. Our family of Triton GEMMs matches NVIDIA cuBLAS in all tested cases on Blackwell and Hopper. For realistic LLM shapes with fused epilogues, it matches or exceeds torch.compile performance. Second, we enforce balanced-tree reduction during Triton lowering and introduce a data-layout optimization that brings 19 of 27 kernels on GB300 and H100 within 10 percent of free-order performance. Third, we develop sound static checkers for bitwise equivalence between compiled GPU kernels, including the first checker spanning NVIDIA PTX and AMD GCN. Integrated into Triton's autotuner, the checker restricts search to a single bit-equivalence class.
Primary: Georgia Institute of Technology
All Institutions: Georgia Institute of Technology, Meta
The paper presents a comprehensive framework for achieving bitwise reproducibility in GPU kernels through black-box reconstruction, compiler enforcement, and static verification. It makes significant contributions to the field of ML systems by providing practical tools and theoretical insights to ensure numerical determinism, which is increasingly critical for reliable machine learning deployments.
The paper introduces a rigorous theoretical framework for characterizing bitwise behavior in GPU kernels, specifically focusing on GEMM and reductions. The core contribution is the "GEMMDesc" descriptor, which formally captures the parameters determining floating-point reduction order (e.g., split-K cuts, instruction-level accumulation semantics). The methodology for black-box reconstruction of closed-source libraries (cuBLAS) is particularly innovative, using numerical inference with specific extreme values to reverse-engineer the internal arithmetic structure without access to source code. The implementation of a sound static equivalence checker for PTX and AMDGCN assembly is a significant technical achievement, providing a formal guarantee that two compiled kernels are bitwise equivalent, which is rare in compiler research.
The evaluation is extensive and rigorous. The authors demonstrate 100% bitwise match with cuBLAS across 110,813 random shapes on multiple GPU architectures (GB300, GB200, H100). The performance analysis is nuanced, showing that while enforcing bitwise determinism typically costs performance, their layout optimizations recover most of this cost, and in some fused kernel scenarios, the bit-exact implementation actually outperforms the baseline due to reduced launch overhead. The soundness of the static checker is validated on over 51,000 configurations, showing high precision in partitioning equivalent kernels.
The paper provides detailed descriptions of the compiler modifications and the static checker algorithm. However, as it is an arXiv preprint, the code availability is not explicitly confirmed in the text provided. The reliance on specific hardware (GB300, H100) and closed-source libraries (cuBLAS) may limit immediate reproducibility for all readers, but the methodology is clearly described.
The work is heavily focused on GEMM and standard reductions; complex kernels like Flash Attention are only theoretically analyzed. The black-box reconstruction is specific to cuBLAS versions tested and may not generalize to other libraries or future hardware without re-derivation. The static checker, while sound, has a performance overhead and may not scale to arbitrarily complex control flows without further optimization.
This work has high potential impact on the ML systems community, particularly for teams requiring strict reproducibility in training and inference. It provides tools to enforce determinism without significant performance loss, addressing a critical pain point in distributed training and RLHF pipelines. The static equivalence checker could become a standard tool in compiler verification suites. The paper presents a comprehensive framework for achieving bitwise reproducibility in GPU kernels through black-box reconstruction, compiler enforcement, and static verification. It makes significant contributions to the field of ML systems by providing practical tools and theoretical insights to ensure numerical determinism, which is increasingly critical for reliable machine learning deployments.