The authors formulate imitation and goal reaching as one goal-conditioned reinforcement-learning problem.
One Human Demonstration Can Seed Adaptable Dexterous Robot Skills
AdaDex combines motion tracking and goal-conditioned reinforcement learning so robot hands can preserve demonstrated strategies while adapting them to new object poses.
Big Tech
Yuxuan Kuang · Jiashun Wang · Yixuan Sun · Yufei Ye · Sungjae Park · Umar Iqbal · +2 more
Carnegie Mellon University · NVIDIA
Research Digest··3 min read
Kuang et al.
§