| Presentation | 2003/10/17 Optimization of observed nominal trajectories : Application to stand-up and biped walking movements Jun MORIMOTO, Kenji DOYA, Christopher G. ATKESON, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In imitation learning framework, because a learner can not detect teacher's policy or exact torque output directly from observed trajectories, the learner can not easily generate target movements by only observing trajectories of teacher's motions. In such a case, the learner needs to optimize the movements based on observed trajectories with considering own dynamics. In this study, we propose to use hierarchical reinforcement learning (HRL) and differential dynamic programming (DDP) to optimize trajectories. As a first example, we apply HRL to stand-up task using 2-ioint 3-link robot model. We show that by using HRL in imitation learning framework, the learner acquire the stand-up movement faster than without using observation of teacher's trajectories. As a second example, we apply DDP to biped walking task using 5-link biped robot model. We show that the learner using robust DDP can generate low-torque and robust biped walking trajectories than a PD servo controller. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | reinforcement learning / differential dynamic programming / stand-up / biped locomotion / imitation learning |
| Paper # | PRMU2003-135,NC2003-66 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 2003/10/17(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Neurocomputing (NC) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Optimization of observed nominal trajectories : Application to stand-up and biped walking movements |
| Sub Title (in English) | |
| Keyword(1) | reinforcement learning |
| Keyword(2) | differential dynamic programming |
| Keyword(3) | stand-up |
| Keyword(4) | biped locomotion |
| Keyword(5) | imitation learning |
| 1st Author's Name | Jun MORIMOTO |
| 1st Author's Affiliation | ATR Computational Neurscience Labs() |
| 2nd Author's Name | Kenji DOYA |
| 2nd Author's Affiliation | ATR Computational Neurscience Labs:CREST, JST |
| 3rd Author's Name | Christopher G. ATKESON |
| 3rd Author's Affiliation | The Robotics Institute, Carnegie Mellon University |
| Date | 2003/10/17 |
| Paper # | PRMU2003-135,NC2003-66 |
| Volume (vol) | vol.103 |
| Number (no) | 392 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |