Event
OneDemo: Learning Policies and Generating Data from a Single Demonstration
1 reports · 1 sources
Ying Qian and co-authors released a paper proposing the OneDemo reinforcement learning framework: it edits a single successful demonstration into trajectories adapted to new task configurations, and the learned editor serves both as the task policy and as a reusable demonstration generator. According to the paper, on 15 Meta-World tasks, OneDemo achieves a 92.7% success rate using an average of 53.7k environment interactions; on 9 DexMimicGen tasks, the generated demonstrations have an average success rate of 98.4%, and BC-RNN policies trained on the generated data achieve an average success rate of 99.6% at the same data scale. All of the above numbers are results reported by the authors in the paper.
Report timeline
- OneDemo: Learning policies and generating data from a single demonstration
OneDemo proposes a reinforcement learning framework that edits a single successful demonstration into trajectories suited to new task configurations; the learned editor serves both as a task policy and as a reusable demonstration generator. On 15 Meta-World tasks, it reaches a 92.7% success rate using an average of 53.7k environment interactions; on 9 DexMimicGen tasks, it generates demonstrations with an average success rate of 98.4%, and BC-RNN policies trained on the generated data reach an average success rate of 99.6% at the same data scale.
arXiv cs.RO ·