研究実績の概要 |
This third fiscal year of the project we achieved tasks and goals from different work-packages (WP). In WP1 that consists in monitoring human subjects in contact and estimating contact forces, we have obtained the following results. We developed multi-contact taxonomies and study of contact invariants. We witnessed that this idea was also studied by the team of Professor Tamim Asfour, at KIT Germany and they published a journal paper on this topic. Deep learning of multi-contact planning strategies was intensively studied. After trying couple of approaches without sound success, we decided to study reinforcement learning technique to try to generate learning of contacts from simulations. In WP2 we have already tackled almost all the subjects. We enhanced the planner and the controller with sliding contacts. The issue here is that sliding in multi-contact requires controlling both tangential forces and the center of mass positioning in a very precise way. We have for now formulated in 2D the positioning of the center of mass depending on the contact forces we aim to target on each contact. For theoretical closed-loop stability and robustness of Quadratic Programming (QP) of multi-contact controllers, the formulation and integrated Simultaneous Localization and Mapping (SLAM) offered very good results. In WP3 we addressed compliance and force regulation, and integration and assessment have been achieved within various demonstrators using the humanoid platform HRP-2Kai and HRP-4. Overall, all the objectives of this fiscal year have been met beyond the expectation.
|