Event
EPCC Combines Slow Demonstrations with Fast Trial-and-Error to Achieve Faster-Than-Demonstration Execution
1 reports · 1 sources
Seunghoon Cho and co-authors propose the Expert-Play Contouring Control (EPCC) method, whose goal is for robot manipulation to execute faster than expert demonstrations. Their approach uses slow expert demonstrations to train a latent trajectory generator, reparameterizing predictions into a time-independent task-progress contour; it then uses fast non-expert trial-and-error data to train a world model that predicts the outcomes of fast actions. At deployment, a planner optimizes actions along this contour and penalizes deviations. The authors report that, averaged over three visuomotor manipulation tasks, EPCC achieves 2.0 times the throughput of imitation learning (IL) baselines; on interaction-sensitive object dynamics tasks, it reaches 2.2 times that of the strongest acceleration baseline.
Report timeline
- EPCC: Faster-than-Demonstration Planning with Slow Expert Demonstrations and Fast Play
The authors propose Expert-Play Contouring Control (EPCC), which combines slow expert demonstrations with fast, non-expert play to achieve execution speeds faster than the demonstrations. Expert demonstrations train a latent trajectory generator whose predictions are reparameterized into a time-independent contour of task progress, while play data trains a world model of fast-action outcomes. At deployment, the planner optimizes actions along this contour and penalizes deviation from it.
arXiv cs.RO ·