Accelerating whole-body motion generation using regression of the torso posture of a humanoid robot

Abstract

In everyday environments such as kitchens, humanoid robots require to have a large workspace. For example, a robot needs to grasp a dish in lower shelves while hunkering down. Furthermore, because the environment is dynamic, solving inverse kinematics for the whole body in real time is necessary. We propose to solve inverse kinematics in real time by splitting it into simpler problems. Given the target configurations of both hands as input, we calculate the orientation of the torso using a regressor and calculate the final position of the Center of Mass (CoM) from the reachability of the arms. We obtain the trajectory of the torso orientation and the CoM using interpolation. In each control step, we first compute the joint angles of the lower body from the CoM position, feet position, and torso orientation. Next, we calculate the joint angles of both arms. In the experiments, we apply the proposed method to the humanoid robot HRP-4 for the task of reaching low-height positions while hunkering down. The proposed inverse kinematics solver is ten times faster than the numerical solution.