Data and Learning Where it Matters for Contact-Rich Manipulation
Robotics
Summary
The authors found that training robots end-to-end on large datasets can still fail at precise tasks because the data collection isn't focused enough. They suggest collecting detailed data only during the tricky parts where the robot touches or manipulates objects and using simpler planning methods for easy movements. They created a system that gathers this critical data automatically and trains the robot offline, without needing a skilled human or updates during use. Their approach worked very well on several real tasks, even outside the usual conditions where other methods often fail.
Authors
Oliver Hausdörfer, Linus Schwarz, Gabor Marko, Christian Dietz, Timo Class, Luka Hofer, Jim Yun-Jin Li, Johannes Hechtl, Ralf Römer, Angela P. Schoellig
Abstract
Learned policies trained end-to-end on large datasets often remain brittle in high-precision tasks and struggle with generalization. We find that these limitations largely stem from a lack of structure and focus in data collection. Our key insight is to leverage dense data collection only for the critical segment of contact-rich tasks and to rely on traditional planning during simple free-space motion. We propose an automated data-collection scheme in combination with offline deep reinforcement learning for the critical segment of the task, eliminating reliance on a teleoperator's skill and on online policy updates. Across four challenging real-world tasks, using only 2 to 2.5 hours of autonomous data collection, we achieve an average success rate of 96%, compared to the strongest baseline at 55%. Notably, performance remains high in out-of-distribution scenarios where end-to-end approaches struggle. Our results pave the way for targeted data collection for contact-rich tasks and for high success rates in precision applications.