Point-cloud multi-contact planning for humanoids: Preliminary results

Abstract

We present preliminary results in porting our multi-contact non-gaited motion planning framework to operate in real environments where the surroundings are acquired using an embedded camera together with a depth map sensor. We consider the robot to have no a priori knowledge of the environment, and propose a scheme to extract the information relevant for planning from an acquired point cloud. This yield the basis of an egocentric on-the-fly multi-contact planner. We then demonstrate its capacity with two simulation scenarios involving an HRP-2 robot in various environment before discussing some issues to be addressed in our quest to achieve a close loop between planning and execution in an environment explored through embedded sensors.