Workshop on Continual Robot Learning from Humans

About
Humans excel at adapting to ever-changing environments, learning from one another, and continuously refining their knowledge and skills over time. The ability to learn continually from human interactions—either through direct demonstrations, communications, proactive feedback, collaboration, or shared norms and values—has been fundamental to human progress. As robots become more integrated into our daily lives, incorporating this adaptive learning in robotics has the potential to revolutionize how they function in dynamic environments, and more importantly, in human society.

How can we develop robots that continually learn and evolve through human interaction? To address this question, the Workshop on Continual Robot Learning from Humans aims at bridging the gap between human adaptive learning and robotic capabilities. The workshop will explore a range of topics, including human-in-the-loop learning, long-term human-robot partnerships, learning societal norms and values, and fostering robot social intelligence. We invite contributions that introduce new theoretical frameworks, present experimental findings, or showcase applications that advance the field of continual learning from humans. Our goal is to tackle emerging challenges and present innovative frameworks that enable robots to learn, adapt, and thrive through ongoing human engagement—fostering an interdisciplinary dialogue that pushes the boundaries of robot learning.

Schedule

Time (PST)
8:50 am - 9:00 am	Organizers Introductory Remarks
9:00 am - 9:30 am	Yonatan Bisk Humans Know Things, Humans Get in the Way Abstract As language models become increasingly pervasive, how do they handle embodied contexts? How well are they calibrated? What's next?!
9:30 am - 10:00 am	Mahi Shafiullah Generalization as a Matter of Perspective Abstract While we are seeing increasingly impressive robot demos and highly capable embodiments come out, we aren't quite at the point where we can have a robot butler in every home. Why is that? In this talk, I will argue that to make advances on this problem we will have to distinguish between interpolation and extrapolation problems in learning. Then, I will talk about how we can cast the problem of robot learning from humans as an interpolation problem with a change of perspective from humans to robots - quite literally - by using handheld tools and an iPhone. I will talk about different approaches that unlock solving problems in novel environments right out of the box following this principle of solving robot problems from the robot perspective. Finally, I will talk about some future challenges that we will have to address, such as dexterity and long horizon, and some solutions that we have worked on.
10:00 am - 10:30 am	Homanga Bharadhwaj Observational Learning for Manipulation via Visual Imitation of Humans Abstract Robots learning to interact with the world by directly observing humans go about their daily life, has been a dream in AI for decades. In this talk, I will discuss recent advances in learning such robotic manipulation zero-shot primarily from large-scale web videos, and also enabling robots to follow in-context human demonstrations one-shot in novel scenarios. Finally, I'll conclude with a reality check about how far we are from the dream, and discuss open challenges.
10:30 am - 11:00 am	Break and Poster Session
11:00 am - 11:30 am	Yilun Du Continual Robot Learning with Compositional Generative Models Abstract Robot learning policies are typically large, monolithic models that are difficult to adapt. In this talk, I'll illustrate how by building policies in in a compositional manner, we can construct systems that can quickly and continually learn from people. I'll first illustrate how we can use inference-time composition to flexibly adapt policies to human constraints. I'll next illustrate how we can use compositional models with language models to quickly learn to set up scenes together. Finally, I'll further illustrate how compositional generative models enable us to construct an inverse generative modeling procedure which allows us to flexibly learn new tasks.
11:30 am - 12:00 pm	Chelsea Finn You Should Talk to You Robot Abstract To be determined.
12:00 pm - 12:30 pm	Maja Mataric HRI to AGI: How Embodied Interaction and Learning Shape Intelligence and Drive the Future of AGI Abstract To be determined.

Best Paper Award

ROSETTA: Constructing Code-Based Reward from Unconstrained Language Preference
Sanjana Srivastava, Kangrui Wang, Yung-Chieh Chan, Tianyuan Dai, Manling Li, Ruohan Zhang, Mengdi Xu, Jiajun Wu, Li Fei-Fei

Spotlight Papers

DemoDiffusion: One-Shot Human Imitation using pre-trained Diffusion Policy
Sungjae Park, Homanga Bharadhwaj, Shubham Tulsiani
Guiding Data Collection via Factored Scaling Curves
Lihan Zha, Apurva Badithela, Michael Zhang, Justin Lidard, Jeremy Bao, Emily Zhou, David Snyder, Allen Z. Ren, Dhruv Shah, Anirudha Majumdar

Papers

DexWild: Dexterous Human Interactions for In-the-Wild Robot Policies
Tony Tao, Mohan Kumar Srirama, Jason Jingzhou Liu, Kenneth Shaw, Deepak Pathak
GZeroMimic: Distilling Robotic Manipulation Skills from Web Videos
Junyao Shi, Zhuolun Zhao, Tianyou Wang, Ian Pedroza, Amy Luo, Jie Wang, Yecheng Jason Ma, Dinesh Jayaraman
DEXOS: Hand Exoskeleton System for Learning Robot Dexterous Manipulation from Human In-The-Wild
Hao-Shu Fang, Arthur Hu, Branden Romero, Edward H Adelson, Pulkit Agrawal
Improving Reward Learning by Estimating Annotator Expertise
Pavel Czempin, Rachel Freedman, Ellen Novoseller, Vernon Lawhern, Cameron Allen, Erdem Biyik
A Generalizable Gripper Interface for Hardware-Agnostic Policy Learning in Robotic Manipulation
Alejandro Gonzalez Hernandez, Mario Cavero-Vidal, Alberto San-Miguel-Tello, Oscar Palacin
Interpretable Human-in-the-Loop In-Context Preference Learning Via Preference Boundaries
Valerie K. Chen, Julie Shah, Andreea Bobu
Multi-RAG: A Multimodal Retrieval-Augmented Generation System for Adaptive Video Understanding
Mingyang Mao, Mariela M. Perez-Cabarcas, Xiaomin Lin, Utteja Kallakuri, Nicholas R Waytowich, Tinoosh Mohsenin
PHIRL: Progress Heuristic for Inverse Reinforcement Learning
HANG YU, James Staley, Shijie Fang, Wenchang Gao, Reuben M Aronson, Elaine Short
Masked Inverse Reinforcement Learning for Language Conditioned Reward Learning
Minyoung Hwang, Alexandra Forsey-Smerek, Andreea Bobu
Causally Robust Preference Learning with Reasons
Minjune Hwang, Yigit Korkmaz, Daniel Seita, Erdem Biyik
From imitation to emulation: A developmental framework for continual robot learning
Chaolan Lin
GRIM: Task-Oriented Grasping with Conditioning on Generative Examples
Shailesh, Alok Raj, Nayan Kumar, Priya Shukla, Andrew Melnik, Michael Beetz, Gora Chand Nandi
HuDOR: Bridging the Human to Robot Dexterity Gap through Object-Oriented Rewards
Irmak Guzey, Yinlong Dai, Georgy Savva, Raunaq Bhirangi, Lerrel Pinto

Call for Papers
We invite original contributions presenting novel ideas and research on continual robot learning from humans. We particularly encourage work emphasizing interdisciplinary approaches and practical applications.

Important Dates

Event	Date
Submission Deadline	~~May 19th, 2025~~ May 22nd, 2025 (23:59 AoE)
Notification	June 7th, 2025 (23:59 AoE)
Camera Ready	June 14th, 2025 (23:59 AoE)
Workshop	June 21st (9:00am - 12:30pm), 2025

Submission Guidelines

Submissions should follow the RSS template, which is available either in LaTeX or Word format. The recommended paper length is 4 pages excluding references. However, any paper that is between 2 and 6 pages, excluding references, will be reviewed for inclusion in the workshop program. All papers must be submitted as anonymized PDFs for double-blind review via OpenReview.

Accepted papers will be presented during the workshop (either in-person or remotely) and featured in a poster session. The proceedings will be treated as non-archival, allowing future conference or journal submissions.

Paper Topics

Relevant topics include, but are not limited to:

Human-in-the-loop methods and feedback-driven robot learning
Long-term human-robot collaboration and partnerships
Adapting to evolving human norms, values, and social context
Integrated Human Feedback to Boost Robot Safety and Performance
Proactive Feedback Strategies for Efficient Robot Learning
Dynamic Models of Human Behavior in Evolving Interactions
Frameworks for Continuous Social Learning and Robust Memory Retention
Adaptive Assistance Systems for Long-Term Human Support
Benchmarks for Evaluating Enduring Human-Robot Collaborations
Balancing Multiple and Conflicting Human Preferences
Mechanisms for Robots to Interpret and Adapt to Changing Social Norms
Pluralistic Human Values for Improved Robot Social Acceptability
Theory of Mind Applications in Robots for Behavior Interpretation and Prediction
Social Reasoning Approaches in Complex Environments
Pedagogical Methods for Enhancing Knowledge Transfer in Robots