Systems like humanoid robots are both underactuated and hyper-redundant. They can move only through their interactions with the environment, i.e. contacts, and have high-dimensional configuration spaces. The need to look for contacts in order to ensure the locomotion implies that any object in the environment can be considered as a possible support. This conflicts with the classical hypothesis in motion planning that considers the objects as obstacles the robot needs to avoid. In the humanoid robot context, objects have a double status : obstacle and support. In this document, we aim at building a motion planner that can handle this double property. We propose a method which works directly in the contact space. We first introduce tools to link the contact space and the configuration space : a posture generator, allowing to project onto submanifolds of the configuration space, and a new kind of bounding volume, called STP-BV, that can be used to introduce collision avoidance constraints within the posture generation. We then describes the global architecture and the various modules of our contact planner, and conclude by showing the result obtained in simulation and on a real robot HRP-2.
