Robotics

Aditi, Niket Agarwal, Arslan Ali et al.

This work provides a scalable, general-purpose backbone for embodied agents by unifying multiple modalities into a single framework, which is a significant step for Physical AI research.

Physical Intelligence, Bo Ai, Ali Amin et al. · mit

For roboticists, π0.7 provides a generalist model that reduces the need for task-specific fine-tuning, enabling broad applicability across platforms and tasks.

Jinghui Lu, Jiayi Guan, Zhijian Huang et al.

For autonomous driving systems requiring real-time decision-making, OneVL provides a method to achieve high accuracy without the latency overhead of autoregressive reasoning.

Tianle Zhang, Zhihao Yuan, Dafeng Chi et al.

This addresses the challenge of insufficient data diversity and poor cross-embodiment generalization for robotic manipulation, representing a novel method rather than an incremental improvement.

Qiuyue Wang, Mingsheng Li, Jian Guan et al.

This work addresses the fragmentation in embodied AI by proposing a unified model that generalizes across diverse tasks and robot platforms, reducing the need for specialized models.

Bohan Hou, Gen Li, Jindou Jia et al.

For researchers in robot learning, this survey organizes a rapidly growing but fragmented field, offering a unified perspective on world models and their applications.

Yu Shang, Yinzhou Tang, Yiding Ma et al.

This benchmark addresses the need for comprehensive evaluation of embodied world models, which is crucial for researchers developing multimodal, interactive, and real-world-capable AI agents.

Aarti Basant, Amlan Kar, Despoina Paschalidou et al. · nvidia

This work addresses the critical bottleneck of safe evaluation of autonomous driving policies in long-tail scenarios by providing a scalable, reactive simulation environment.

Davis Rempe, Mathis Petrovich, Ye Yuan et al.

This addresses the need for scalable, high-quality human motion data for applications in robotics, simulation, and entertainment, representing a significant advancement over previous limited datasets.

Siyin Wang, Junhao Shi, Zhaoyang Fu et al.

For researchers in embodied AI, this survey offers the first systematic framework to understand and compare WAM approaches, clarifying architectural trade-offs and future directions.

Zongzheng Zhang, Jingrui Pang, Zhuo Yang et al.

It addresses the lack of open-source VLA systems for high-degree-of-freedom bimanual dexterous manipulation, enabling broader research in embodied AI.

Kaidong Zhang, Jian Zhang, Rongtao Xu et al.

This work addresses the problem of expensive real-time robot control for researchers and practitioners by providing a more efficient and transparent solution, though it is incremental in optimizing existing methods.

Jiazhi Yang, Kunyang Lin, Jinwei Li et al.

For robotic manipulation, RISE enables safe and scalable reinforcement learning without physical interaction, significantly improving robustness in contact-rich tasks.

Heng Zhou, Li Kang, Yiran Qin et al.

This addresses the problem of collaborative spatial reasoning for embodied AI systems, offering a principled foundation for learning world-centric scene understanding from ego-centric observations, though it appears incremental as it builds on existing methods like reinforcement learning and vision-language models.

Junli Ren, Yinghui Li, Kai Zhang et al.

This work addresses the challenge of dynamic humanoid interaction tasks for robotics, advancing beyond prior systems that relied on external sensing and decoupled control.

Haoquan Fang, Jiafei Duan, Donovan Clay et al.

For robotics researchers and practitioners, this work provides a fully open, high-performing VLA model with practical deployment considerations (latency, hardware cost), though it is an incremental improvement over existing VLA approaches.

Yongkang Li, Lijun Zhou, Sixu Yan et al.

It addresses a critical bottleneck in autonomous driving systems by improving model capabilities for perception and reasoning, though it appears incremental as it builds on existing VLA frameworks.

Open-H-Embodiment Consortium, Nigel Nelson, Juo-Tung Chen et al.

For medical robotics researchers, this dataset and models address the data bottleneck hindering foundation model development, providing critical infrastructure for robot learning and world modeling.

Shuanghao Bai, Meng Li, Xinyuan Lv et al.

This addresses the problem of unstable and inefficient whole-body control for humanoid robots, enabling better manipulation in fast-reaction and long-horizon scenarios.

Charles Xu, Jost Tobias Springenberg, Michael Equi et al.

For robotics practitioners, this enables rapid fine-tuning of large VLAs to achieve precise and fast manipulation skills with minimal real-world interaction.