Footstep planning for a humanoid robot using visually reconstructed terrain

Abstract

This master’s thesis presents a footstep planning for a humanoid robot with perception and pro cessing of the visual information. Our approach consists of two main parts. First, the processing and the segmentation of the data from RGB-D cameras identify the ground and other planes where the robot can possibly step on. Then, methods such as pose estimation and 3D reconstruction are used to build a map of the environment. Second, we study and integrate the reachability map, which represents the kinematically feasible configurations to a path planning algorithm. The method has been implemented for the DLR humanoid robot TORO on the simulation using two walking methods. The first method uses the Divergent Component of Motions on a Linear Inverted Pendulum model with a feedback controller to generate the trajectory of the robot. The second method uses the multi-contact planner and controller approach which computes the center of mass position and the trajectories of the end-effectors. The results on simulation demonstrate the robustness and the suitability of our footstep planning approach.