Unsupervised skill discovery approach using automatically generated tasks and asymmetric self-play to learn composable robotic manipulation behaviors.
Oct 7, 2024
Multi-critic actor-critic approach that combines multiple reward functions to discover safe, useful robotic manipulation skills.
Feb 1, 2024
End-to-end deep reinforcement learning controller for the open-source Solo12 quadruped robot, validated on real hardware.
Aug 2, 2023
LLM-based approach that automatically converts text-based task descriptions into reward and goal-generation functions for robotic manipulation.
Jun 19, 2023
Safe skill discovery approach for robotic manipulation combining reinforcement learning with safety constraints.
Jan 1, 2023
Novel training approach using reachable manifold and inverse mapping for improved robotic manipulation.
Jan 1, 2021