Event
Study: Low-resolution proximity sensors are more beneficial for humanoid robot collision avoidance learning
1 reports · 1 sources
Authors including Carson Kohlbrenner published a study that uses whole-body proximity sensors as an inductive bias to train humanoid robots to learn reflexive collision-avoidance behaviors. The authors state that under deliberately constrained sample and compute budgets, low-resolution lossy sensors improve learning outcomes more than high-resolution full-coverage sensors. The study compared eight candidate representations in a reinforcement learning environment, including latent-space compression, future contact prediction, lossy degradation, and privileged state estimation, training end-to-end more than 200 humanoid robots playing a simplified version of dodgeball. The authors also found that predicting when and where contact will occur instead reduced learning sample efficiency.
Report timeline
- Move, don't think: Sample-efficient collision avoidance learning with lossy proximity sensors
This paper investigates whole-body proximity sensors as an inductive bias for learning reflexive collision-avoidance behavior, finding that low-resolution lossy sensors were a stronger bias than high-resolution full coverage. The study compared eight candidate representations in a reinforcement learning environment, including latent compression, future contact anticipation, lossy corruption, and privileged state estimation, and trained over 200 humanoid robots end-to-end to play a simplified version of dodgeball under intentionally constrained sample and compute budgets. The authors also found that anticipating when and where contact occurs actually reduced learning sample efficiency.
arXiv cs.RO ·