one-shot high-fidelity imitation | Jiao's blog
Opened this issue · 0 comments
Hotwaterman commented
https://hotwaterman.github.io/post/one-shot-high-fidelity-imitation/
人类是高保真模仿的专家，只需要一次尝试就能模仿一个示范动作。这篇paper引入了off-policy RL算法 MetaMimic，meta mimic可以学习1）不同新技能的高保真one-shot模仿的策略，2）使agent比演示者更有效...