Model predictive control for leg/wheel mobile robots using partitioned model

Abstract

Leg/Wheel mobile robots can achieve adaptive motions for various environments using legs or wheels. However, computational complexity of optimization like model predictive control (MPC) is increased with the number of degree of freedom on joints and wheels. In this research, we propose to partition a model of a leg/wheel mobile robot to decrease computational complexity. A model of leg/wheel mobile robot is represented by plural symmetric model for a leg composed of joints and a wheel. The robot is controlled by optimizing each leg sequentially using partitioned model at every control cycle. Partitioned models are connected by constraints to maintain body shape. While the configuration of a single leg is optimized, the robot can achieve concerted motion using predicted state of the other legs in the constraint. In this paper, we conduct tracking trajectory control of wheel position using partitioned model. We choose a difficult trajectory for the robot which includes rotational and translational motion to verify the performance of tracking trajectory. We confirm that the tracking performance is not deteriorated by experiment verification using an actual robot, while the concerted motion is achieved even if the behavior of each leg is optimized sequentially. Moreover, we verify that the computational complexity is decreased by the proposed method.