Online DCM Trajectory Generation for Push Recovery of Torque-Controlled Humanoid Robots

TL;DR

This paper introduces an efficient online method for humanoid robot push recovery that plans DCM trajectories and adjusts steps in real-time, preventing falls during walking under external pushes.

Contribution

It proposes a novel step adapter integrated with DCM planning, formulated as a quadratic programming problem for real-time push recovery in torque-controlled humanoid robots.

Findings

Successfully prevents robot from falling under external pushes

Operates in real-time during walking at 0.28 m/s

Validated on the iCub humanoid robot with external forces up to 150N

Abstract

We present a computationally efficient method for online planning of bipedal walking trajectories with push recovery. In particular, the proposed methodology fits control architectures where the Divergent-Component-of-Motion (DCM) is planned beforehand, and adds a step adapter to adjust the planned trajectories and achieve push recovery. Assuming that the robot is in a single support state, the step adapter generates new positions and timings for the next step. The step adapter is active in single support phases only, but the proposed torque-control architecture considers double support phases too. The key idea for the design of the step adapter is to impose both initial and final DCM step values using an exponential interpolation of the time varying ZMP trajectory.This allows us to cast the push recovery problem as a Quadratic Programming (QP) one, and to solve it online with state-of-the-art optimisers. The overall approach is validated with simulations of the torque-controlled 33 kg humanoid robot iCub. Results show that the proposed strategy prevents the humanoid robot from falling while walking at 0.28 m/s and pushed with external forces up to 150 Newton for 0.05 seconds.