Get Started
Home
Topics
Search
Library
Research questionHow can web videos provide relevant human demonstrations for long-tail dexterous robot manipulation?Robot demonstrations are expensive and sparse for the long tail of manipulation tasks. Although web videos are abundant, viewpoint changes, scene variation, and actor occlusions make relevant demonstrations difficult to identify for robot learning.
Computer Vision
Image & Video Processing
Information Retrieval
Machine Learning
Robotics
Latest papersRecent research connected to this question, newest first.RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation LearningThe source studies RoboTok, which retrieves human manipulation videos using 3D hand-trajectory representations and evaluates retrieval quality alongside downstream dexterous robot policy performance. Evidence is limited to the reported retrieval benchmarks and downstream task results.research paper · Sep 2, 2026
Related questions
How can dexterous robot hands generalize a manipulation skill from one demonstration to varied objects despite sim-to-real gaps?How can robots learn robust, generalist bimanual household manipulation from limited human demonstrations?How can robots safely learn dynamic manipulation skills online despite sim-to-real mismatch?How can robot perception encode action-relevant scene dynamics to improve manipulation generalization?