Robots operating in real-world environments often learn from human demonstrations, preferences, or rewards. However, in many deployment settings, people act as supervisors rather than teachers: they monitor robot behavior and intervene only when necessary. These corrections and interventions provide rich information about task objectives, constraints, and human expectations, while requiring less effort than continuous supervision.
Learning from Corrections (LfC) has emerged as a promising paradigm for interactive robot learning, but many fundamental questions remain:
What constitutes a correction or intervention?
What information is contained in corrective behavior?
Why and when do people intervene?
How can robots learn from noisy and imperfect corrections?
How can correction-based learning scale to real-world robotic systems?
This workshop brings together researchers from robotics, machine learning, human-robot interaction, and related application domains to discuss the foundations, challenges, and future directions of Learning from Corrections.
Topics
Definitions, taxonomies, and terminology for corrections and interventions
Explicit versus implicit corrective feedback
Multimodal correction signals
Datasets and data collection methodologies
Standardized representations and annotations
Trust, expectations, and perceived robot competence
Human supervision strategies
Mental models and human-robot alignment
Intervention timing and prediction
Individual differences and personalization
Learning reward functions, policies, and constraints
Modeling uncertainty and noisy feedback
Learning from explicit and implicit correction signals
Adapting methods from demonstrations and preferences
Evaluation metrics and benchmarks
Shared autonomy and supervisory control
Assistive robotics
Foundation-model robotics
Human-AI oversight
Safety-critical systems
Industry deployment of intervention-based learning
Call for Papers
We invite submissions on all aspects of Learning from Corrections and Interventions, including:
Correction representations
Human factors
Interactive robot learning
Shared autonomy
Preference learning
Foundation-model robotics
Real-world deployment
We welcome completed research papers, works-in-progress, datasets, benchmark proposals, and position papers.
Invited Speakers
Andreea Bobu
Human feedback
Shared task representations
Value alignment
Siddharth Karamcheti
Scalable robot learning
Foundation models
Real-world deployment
Dylan Losey
Physical human corrections
Interactive robot learning
Learning from interventions
Emmanuel Senft
Human-robot interaction
Human teaching behavior
Human guidance and feedback
Schedule
8:30--8:45: Invited Speaker #1
8:45--9:15: Lightning Talks #1
9:15--9:30: Invited Speaker #2
9:30--10:00: Lightning Talks #2
10:00--10:30: Poster Session #1
10:30--11:00: Break
11:00--11:30: Poster Session #2
11:30--11:50: Breakout Session
11:50--12:05: Breakout Presentations
12:05--12:20: Invited Speaker #3
12:20--12:30: Closing Remarks
Organizers
Anjiabei Wang — Yale University
Tesca Fitzgerald — Yale University
Erdem Biyik — University of Southern California
Yuchen Cui — UCLA
Contact
In case of any questions regarding the workshop, please email anjiabei,wang@yale.edu.