Résumé
We investigate the problem of planning sequences of contacts that support acyclic motion for a humanoid robot on uneven floors or in highly constrained environments. We propose a method that combines two basic techniques for motion generation: planning and optimization. The planning part is guided by a potential field function that grows a tree whose nodes are sets of possible contacts; the optimization part consists in a posture generator that checks the feasibility of contactsset candidates under constraints such as static stability, joint limits, non-desirable collisions, etc. The difference between successive nodes in the tree is exactly one contact (creation or removal). Our planner allows motion-supporting contacts to be made with any parts of each robot's links on any part of the environment: it can be seen as the generalization of motion in which walking is a particular case. Therefore, the meaning of obstacles is different from the classical usage in planning: any part of the environment is at the same time a potential support (for some parts of the robot) and an obstacle (for the remaining parts). We demonstrated our planner in various scenarios involving the HRP-2 humanoid robot and conducted several experiments.