Online DNN-driven Nonlinear MPC for Stylistic Humanoid Robot Walking with Step Adjustment

TL;DR

A three-layered architecture that enables stylistic locomotion with online contact location adjustment that combines an autoregressive Deep Neural Network acting as a trajectory generation layer with a model-based trajectory adjustment and trajectory control layers is presented.

Abstract

This paper presents a three-layered architecture that enables stylistic locomotion with online contact location adjustment. Our method combines an autoregressive Deep Neural Network (DNN) acting as a trajectory generation layer with a model-based trajectory adjustment and trajectory control layers. The DNN produces centroidal and postural references serving as an initial guess and regularizer for the other layers. Being the DNN trained on human motion capture data, the resulting robot motion exhibits locomotion patterns, resembling a human walking style. The trajectory adjustment layer utilizes non-linear optimization to ensure dynamically feasible center of mass (CoM) motion while addressing step adjustments. We compare two implementations of the trajectory adjustment layer: one as a receding horizon planner (RHP) and the other as a model predictive controller (MPC). To enhance MPC performance, we introduce a Kalman filter to reduce measurement noise. The filter parameters are automatically tuned with a Genetic Algorithm. Experimental results on the ergoCub humanoid robot demonstrate the system's ability to prevent falls, replicate human walking styles, and withstand disturbances up to 68 Newton. Website: https://sites.google.com/view/dnn-mpc-walking Youtube video: https://www.youtube.com/watch?v=x3tzEfxO-xQ

Figures (6)

• Figure 1: The humanoid robot ergoCub walks forward, sideways, and reacts to an external force with the proposed locomotion architecture.
• Figure 2: The proposed three-layer hierarchical architecture integrating a as a trajectory generation layer and a model-based trajectory adjustment layer.
• Figure 3: Top-view of the walking pattern performed by ergoCub with the trajectory adjustment layer acting as . The robot is pushed twice. The red arrows represent the external pushes.
• Figure 4: Measured upper body joint angles - specifically, torso and left arm - compared with their associated data-driven postural during walking.
• Figure 5: Top-view of the walking pattern performed by ergoCub with the trajectory adjustment layer functioning as a model predictive controller . The robot is pushed three times, as indicated by the red arrows. For the second (B) and third (C) disturbances, occurring at 61.9s and 69.43s respectively, the lateral (local y-coordinate) velocity is displayed. The red dashed line marks the time at which the system is perturbed.