Modeling and MPC-Based Pose Tracking for Wheeled Bipedal Robot

Abstract

In this letter, we propose a model predictive control (MPC)-based robot pose controller for our newly designed wheeled bipedal robot (WBR). The proposed controller uses the virtual model control concept, allowing for wider applicability by ignoring the leg dynamics. By directly incorporating the non-holonomic constraint of the wheels into the dynamic equation, a wheeled rigid dynamic model is proposed to maximize the motion flexibility and minimize the model order. A hierarchical MPC control structure is employed to track the desired pose while considering the non-minimal phase property of WBRs in real time. To enhance the autonomy of the robot, we propose a state estimator that utilizes kinematics and inertial sensor data to provide a high-speed and accurate estimation of the robot's state. Both simulation and real-world experiments demonstrate that the proposed method can track a pose trajectory with lower error than traditional feedback control methods. The effectiveness of the estimator is validated through comparison with motion capture cameras and vision-based odometry.