Data-driven tube model predictive control of autonomous agricultural tractors for cross-slope navigation

Autonomous navigation control of agricultural tractors on cross-slope terrain presents significant challenges due to sustained lateral disturbances. This study introduces an innovative navigation control method for agricultural tractors on cross-slopes, integrating Long Short-Term Memory (LSTM) networks with robust Tube Model Predictive Control (TMPC). The approach utilizes LSTM for real-time slope estimation and disturbance prediction over the prediction horizon. These estimations are used to scale pre-computed ellipsoidal disturbance sets, forming time-varying probabilistic disturbance boundaries for the TMPC framework. We prove probabilistic recursive feasibility and closed-loop stability under this adaptive scheme. This design significantly reduces the conservativeness of TMPC in scenarios with unknown disturbances. Experimental results demonstrate a notable reduction in lateral tracking errors using the LSTM-augmented TMPC, enhancing the navigation accuracy and stability of tractors on cross-slope terrain. This research not only showcases the capability of LSTM in precise slope estimation under complex environments, but also validates the practicality and efficacy of probabilistic robust MPC in agricultural tractor navigation control.