Human-Centered Robot Learning in the Era of Big Data and Large Models
May 19th, 2025 | Atlanta, Georgia, USA | ICRA 2025
Recent advances in training large-scale models on massive datasets have led to the development of foundational models, such as large language models (LLMs) and large multimodal models (LMMs), capable of interacting with humans to perform sophisticated tasks. These models also enable new paradigms for human-robot interaction, facilitating the deployment of robots in human-centered environments to operate alongside humans (e.g., transportation, household tasks, healthcare, and warehouses). For example, robots powered by LLMs and LMMs show promise in following complex human instructions, collaborating with humans, and understanding and adhering to social norms.
However, compared to deploying AI models that interact with humans virtually (e.g., chatbots), deploying those embodied AI agents to interact with humans in the physical world imposes various unique challenges, for example: 1) The models need to be grounded in the physical world, especially in human representations and understanding of the physical world, to facilitate efficient and trustworthy human interaction; 2) The risk of physical harm to humans imposes significantly higher requirements on AI safety and alignment methods; 3) Robot data, especially human-robot interaction data, is much less accessible, compared to the abundance of internet data used for training virtual AI models.
This workshop aims to bring together interdisciplinary experts from related fields such as robot learning, human-robot interaction, natural language processing, cognitive science, and trustworthy AI to gather and discuss these challenges. We envision that this cross-disciplinary exchange will build new pathways for developing and deploying large-scale models in human-centered embodied AI systems. We intend to guide our conversations with the following questions:
- What are the ideal data types for learning embodied interactions, and how can we acquire them efficiently and safely?
- How can we train large models for human-centered robots to ensure robustness across diverse human interaction behaviors?
- What challenges do human-centered robotic applications pose for AI safety and alignment, and how can we align robot behaviors with stakeholder values?
- What new interaction paradigms between humans and robots are enabled by big data and large models? Given these new paradigms, what are the best practices for allocating tasks to robots and people in a collaborative setting?
Speakers
Call For Papers
Submission deadline: April 6th, 11:59pm AOE, 2025 April 7th, 11:59pm AOE, 2025
Review and decision: April 20th, 2025
Camera-ready deadline: May 11th, 2025
We invite submissions that address the fundamental challenges in enabling large-scale human-centered robot learning or demonstrate novel human-centered embodied AI systems enabled by large-scale data and models. The related topics include but are not limited to:
- Foundation models for embodied AI
- Data accessibility for embodied AI
- AI Safety and alignment for embodied AI
- Efficient fine-tuning of pre-trained large models for embodied AI
- Applications in human-robot interaction and collaboration
Submission Guidelines
The review process will be single-blind. Submissions will be evaluated based on originality, technical quality, and relevance to the workshop themes. The review process will not be public; only the accepted papers will be available on the workshop website, given the authors’ consent.
There is no strict page limit; however, we encourage submissions to be within 4-8 pages (excluding references) to facilitate thorough review. Papers should be submitted in PDF, adhering to the ICRA template and our submission guidelines. Also, we encourage authors to submit videos, code, or data in their supplementary material (zip file) or through external services like Github repos.
Additionally, to encourage discussion on the workshop topic, we ask authors to provide 1~2 paragraphs of statements on how their work relates to the workshop theme. This information will help us organize the workshop discussion and better feature the accepted works on the website.
Schedule
|
|
Opening Remarks
Opening Remarks by Workshop Organizers |
Organizers
|
|
|
Keynote Talk 1
Models, Trust, and Unknown Unknowns |
|
|
|
Keynote Talk 2
Data-Driven Approaches for Policy Learning in Household Robots |
|
|
|
Spotlight Talks of Contributed Papers
|
Authors of contributions
|
|
|
Coffee Break and Poster Session
|
|
|
|
Keynote Talk 3
Learning from Intuitive Interactions with Human Teachers |
|
|
|
Keynote Talk 4
(Tentative) Rethinking Social, Task, and Motion Planning with Foundation Models |
|
|
|
Lunch Break
|
|
|
|
Keynote Talk 5
What Can Robot Safety Learn from LLM Safety? |
|
|
|
Keynote Talk 6
Value Alignment and Safety in Human-Robot Interactive and Explainable Learning with Models Big and Small |
|
|
|
Spotlight Talks of Contributed Papers
|
Authors of contributions
|
|
|
Coffee Break and Poster Session
|
|
|
|
Panel Discussion
Panel Discussion with Keynote Speakers |
All Speakers
|
|
|
Closing Remarks
Closing Remarks by Workshop Organizers |
Organizers
|
