A model predictive trajectory tracking control strategy for heavy-duty unmanned tracked vehicle using deep Koopman operator

Abstract

Among the numerous technologies for the heavy-duty unmanned tracked vehicle (HDUTV), trajectory tracking is the key function to support the maneuverability. Unlike Ackermann steering vehicles, HDUTVs are easily affected by disturbances during the steering process, leading to different steering characteristics. The variable steering characteristics pose challenges for precise tracking control. Motivated by this challenge, a high accuracy model predictive trajectory tracking method is proposed to improve the tracking performance of HDUTVs. First, a deep Koopman operator-based tracked vehicle model is established. The proposed learning-based model provides an accurate description of the complex nonlinear dynamics of HDUTVs while maintaining the model linearity. Utilizing the model, the real-time performance of the trajectory tracking process is guaranteed. Second, a trajectory tracking control strategy is established considering the steering characteristic of the HDUTV to improve the tracking performance. Third, the deep Koopman operator-based model is integrated into the model predictive control framework to enhance predictive accuracy while ensuring the real-time performance of the trajectory tracking controller. Finally, the proposed method is validated through simulations and experiments with a full-sized HDUTV. Experiment results indicate that the proposed model enhances predictive ability for vehicle states, with a 59.51 % improvement in the accuracy of the sideslip angle. And the proposed trajectory tracking strategy improves the tracking accuracy by 57.93 %.