Event
Proposing a motion and style prior method for legged locomotion learning
1 reports · 1 sources
Oliver Hausdörfer and other researchers propose an inductive bias method for joint-level torque control of legged robots: learning a latent motion prior from a small amount of expert demonstration data, combined with style rewards for imitation learning. The researchers state that this prior frees the agent from the limitation of the demonstration reward level, significantly improving transfer task performance; after adding style rewards, the agent's behavior can more closely match the experts. Related videos have been made public.
Report timeline
- Learning Action and Style Priors for Joint-Torque Control in Legged Robots
A study on deep reinforcement learning for legged robots proposes using latent actions learned from a small set of expert demonstrations as an inductive bias to help joint-level torque control policies explore more efficiently. This prior allows the agent to not be limited by the demonstration reward level and significantly improves performance on transfer tasks; when combined with style rewards for imitation, it can more closely match expert behavior. Related videos have been made public.
arXiv cs.RO ·