Abstract | ||
---|---|---|
AbstractWe address the longstanding challenge of producing flexible, realistic humanoid character controllers that can perform diverse whole-body tasks involving object interactions. This challenge is central to a variety of fields, from graphics and animation to robotics and motor neuroscience. Our physics-based environment uses realistic actuation and first-person perception - including touch sensors and egocentric vision - with a view to producing active-sensing behaviors (e.g. gaze direction), transferability to real robots, and comparisons to the biology. We develop an integrated neural-network based approach consisting of a motor primitive module, human demonstrations, and an instructed reinforcement learning regime with curricula and task variations. We demonstrate the utility of our approach for several tasks, including goal-conditioned box carrying and ball catching, and we characterize its behavioral robustness. The resulting controllers can be deployed in real-time on a standard PC.1 |
Year | DOI | Venue |
---|---|---|
2020 | 10.1145/3386569.3392474 | ACM Transactions on Graphics |
Keywords | DocType | Volume |
reinforcement learning, physics-based character, motor control, object interaction | Journal | 39 |
Issue | ISSN | Citations |
4 | 0730-0301 | 1 |
PageRank | References | Authors |
0.35 | 0 | 9 |
Name | Order | Citations | PageRank |
---|---|---|---|
Josh S. Merel | 1 | 143 | 11.34 |
Saran Tunyasuvunakool | 2 | 1 | 0.35 |
Arun Ahuja | 3 | 72 | 7.45 |
Yuval Tassa | 4 | 1097 | 52.33 |
Leonard Hasenclever | 5 | 20 | 5.42 |
Vu Pham | 6 | 15 | 2.28 |
Tom Erez | 7 | 1027 | 50.56 |
Greg Wayne | 8 | 592 | 31.86 |
Nicolas Heess | 9 | 1762 | 94.77 |