Humanoid Agents Workshop

About

Building AI agents that behave just like human beings—i.e., Humanoid Agents—has long been a central goal in AI research. Recent advancements have brought us closer to this goal, whether through humanoid avatars in virtual worlds or humanoid robots in the real world.

Humanoid avatar in the virtual world [1]

Humanoid robot in the real world [2]

Developing such agents is an inherently interdisciplinary challenge, with various fields addressing distinct yet interconnected aspects of the problem:

Computer Vision: motion capture and synthesis, human interaction modeling and generation
Computer Graphics: character animation, physics-based simulation, digital humans for mixed reality
Robotics: locomotion, whole-body control, dexterous manipulation, humanoid foundation models
Cognitive Science: autonomous decision-making, human-agent interaction, social cognition

With many exciting advancements across these fields, we believe now is the perfect time to bring together researchers from diverse disciplines for an open and insightful discussion.

This workshop aims to facilitate the exchange of ideas, explore interdisciplinary intersections, and identify key challenges that can spark new insights and innovations. Additionally, it will serve as an opportunity to discuss ethical concerns and societal impacts, and strategies to address them through collective effort.

Call for Papers

We welcome submissions on frontier topics in humanoid agents, including but not limited to the aforementioned areas. Given the interdisciplinary nature of this workshop, we encourage submissions not only from the computer vision community but also from the fields of computer graphics, robotics, cognitive science, etc., as well as from industry practitioners working on related challenges.

Submissions must be anonymized and limited to a maximum of 8 pages for the main content, with no restrictions on the length of references or appendices. Authors are encouraged to use the CVPR template.

We also welcome relevant submissions recently published at other venues (e.g. NeurIPS / ICLR / SIGGRAPH / ICRA), but ask authors to indicate this upon submission. For such submissions, there are no template or page limit requirements. All accepted papers will be presented as posters, with selected papers featured as spotlight talks.

We seek to foster open discussions on, but not limited to, the following topics:

How can real-world human data (e.g., motion capture, internet videos) be utilized to advance humanoid agent behaviors?
Can virtual agents serve as simulations to aid in the development of real-world humanoid robots? What are the current limitations of existing simulation environments? Can improvements in animation quality benefit robot policy learning?
Can we build foundation models for humanoid agents to enable generalizable and adaptable behaviors across diverse tasks and environments? How can traditional control methods be integrated with foundation models?
How can humanoid agents not only exhibit human-like movement but also generate human-like behaviors autonomously?
How can insights from human behavior research (e.g., psychology, cognitive science, neuroscience) inform the development of humanoid agents?
How will humans interact with humanoid agents? What measures can be taken to promote safety, interpretability, and responsible usage in these interactions?