2016 Fiscal Year Annual Research Report
Cutting-edge multi-contact behaviors
| Project/Area Number |
16H02886
|
|---|
| Research Institution | National Institute of Advanced Industrial Science and Technology |
|---|
Principal Investigator |
Kheddar Abder 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 国際客員研究員 (90572082)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
森澤 光晴 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (00392671)
金広 文男 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究グループ長 (70356806)
吉田 英一 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 副研究部門長 (30358329)
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Keywords | 知能ロボティクス / ディジタルヒューマンモデル / 多点接触動作 / 機械学習 / 最適化 |
|---|
| Outline of Annual Research Achievements |
We devised a framework for real-time online and offline retargeting of a human actor motion to a humanoid robot motion involving multi-contact configuration changes between the human/humanoid and their environments. The framework is based on the specification within a multi-contact QP control formulation of tracking tasks for a selected set of body-segments/links, used as contact supports or as locomotion supports on the environment. The framework is applied in an online setting for simultaneous human-robot motion tracking and as a necessary motion information exploited by the tracking algorithm. This approach is validated on a real robot experiment with HRP-4.
We have also proposed a novel technique for unobtrusively estimating interaction forces exerted by human subjects in multi-contact. Our method uses motion capture only, which is particularly challenging, as the knowledge of a given motion only characterizes the resultant force. We collect a large-scale dataset on how humans instinctively regulate interaction forces on diverse multi-contact tasks and motions. The force estimation framework leverages physics-based optimization to reconstruct force distributions that are simultaneously physically realistic.
We are still at the investigation stages for the two issues (2) development of multi-contact taxonomies and study of contact invariants and (3) deep learning of multi-contact planning strategies, to figure out what could be the good features to learn and how the problem is to be formulated.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
For FY2016 we have achieved multi-contact human capture set-up and observations and developed a novel technique for unobtrusively estimating interaction forces, which has been published and obtained an award in an international conference. The research on taxonomy and deep learning is still in progress, but we have already started some subjects and obtained results on contact stability analysis and closed-loop control planned in FY2017 and FY2018 respectively. We then estimate the project is advancing as expected.
|
| Strategy for Future Research Activity |
We will continue the research as planned in the project that is divided in WP1 (FY206) to WP3 (FY2018). There could however be some subjects in each WPs that may delay compared to the plans due to unexpected reasons. In those cases, we identify the topics that can be tackled in the upcoming WPs as much as possible while trying to catch up the remaining work.
|
