Multi-Contact Motion Planning for the Humanoid Robot TORO

Abstract

This master thesis presents a new planning framework for multi-contact interaction of humanoid robots. This frameworks includes a posture generator, which finds the joint configuration of the robot given the position and orientation of the end-effectors and the information about the links of the robot in contact with the environment. The posture generator is a gradient-based inverse kinematics that meets a series of typical constraints for humanoids, such as: stability, torque and joint limits, collisions with itself and with the environment, friction, and singularity avoidance. This thesis explains how each of these constraints are formulated and how they are integrated to the inverse kinematics problem. For some of them, a Quadratic Programming optimization is proposed. In addition, the framework also incorporates a planning algorithm which meets all given constraints. The algorithm is based on a modified version of Bidirectional Rapidly exploring Random Trees, that meets specified constrains and ensures multi-contact interaction of the robot. For the generation of paths of complete tasks involving several contact changes, a tree structure is implemented which foresees possible stagnations and tries to find an alternative path. The planning algorithm proposed here is tested and evaluated using the humanoid robot TORO. Experiments are carried out first in simulation and then with the physical robot. Therefore, the model of the robot and the virtual environment are also described in this thesis. The considered experiment is climbing a three-rungs stairs using a handrail as supporting object, where the robot can make contact. In this manner, the multi-contact capabilities of the planning framework are exhibited.