Online Iterative LQR Controller for Wheeled Bipedal Robot Trajectory Tracking

Wheeled biped balancing robot is an underactuated system with compact actuators and high energy efficiency. Balancing and rotating motions are coupled nonlinearly, posing a major challenge for accurate trajectory tracking. Although yaw motion can be implemented as a task of whole body control and realized by a pole-assignment based controller, model-predictive control (MPC) can significantly reduce trajectory tracking errors. We first derive the system dynamic of robot motion on flat ground. To take into account the dynamic nonlinearity, we deploy iterative LQR as a nonlinear MPC scheme and develop a universal real-time control framework for a nonlinear time-varying system. The proposed system architecture is deployed on a wheeled biped robot platform and tested to track time-dependent B-spine trajectory. Experimental results of physical experiments shows that our online iterative LQR feedback scheme outperforms the baseline controller and shows robustness to disturbances.

I. Introduction

Wheel-legged robots have been gathering increasing attention for their agile maneuverability and terrain traverse ability. By traveling across challenging terrains with high energy efficiency using reductive actuators, wheeled bipedal robots hold significant promise for affordable real-world applications ranging from search and rescue missions to complex environment exploration.