ASE: Large-Scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters
Learns reusable skill embeddings for simulated characters by combining adversarial imitation and unsupervised RL, so new tasks need not start from scratch.结合对抗模仿与无监督强化学习，为仿真角色学习可复用的技能嵌入，新任务不必从零训练。
BibTeX
@misc{peng2022ase,
title = {{ASE: Large-Scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters}},
author = {Xue Bin Peng and Yunrong Guo and Lina Halper and Sergey Levine and Sanja Fidler},
year = {2022},
eprint = {2205.01906},
archivePrefix = {arXiv},
primaryClass = {cs.GR},
url = {https://arxiv.org/abs/2205.01906},
}Locomotion & Humanoids运动控制与人形机器人Planning, Control & Safety规划、控制与安全characterskill-embeddingadversarial