Torque and velocity controllers to perform jumps with a humanoid robot: theory and implementation on the iCub robot

TL;DR

This paper presents two novel control approaches for enabling a humanoid robot to perform jumps, emphasizing the importance of centroidal angular momentum control and joint performance, validated through simulation and real-world experiments on the iCub robot.

Contribution

It introduces a velocity and a torque controller based on optimization techniques for jumping, with implementation and testing on the iCub humanoid robot.

Findings

Both controllers enabled jumping in simulation.

Only the velocity controller succeeded on the real robot.

Controlling centroidal angular momentum is crucial for successful jumps.

Abstract

Jumping can be an effective way of locomotion to overcome small terrain gaps or obstacles. In this paper we propose two different approaches to perform jumps with a humanoid robot. Specifically, starting from a pre-defined CoM trajectory we develop the theory for a velocity controller and for a torque controller based on an optimization technique for the evaluation of the joints input. The controllers have been tested both in simulation and on the humanoid robot iCub. In simulation the robot was able to jump using both controllers, while the real system jumped with the velocity controller only. The results highlight the importance of controlling the centroidal angular momentum and they suggest that the joint performances, namely maximum power, of the legs and torso joints, and the low level control performances are fundamental to achieve acceptable results.