Hierarchical robust motion control for electric-driven tracked vehicles with SAC-based steering-demand generation and torque coordination

Dual-side independently driven electric tracked vehicles have become an important research focus owing to their flexible and efficient propulsion configuration. However, variations in payload and terrain conditions complicate the coordination of longitudinal and steering demands and degrade vehicle-motion robustness. To address these issues, this paper proposes a hierarchical motion-control strategy. In the upper layer, an interpretable rule-based mapping converts the pedal input and current motor speed into a common longitudinal-torque demand, whereas a soft actor–critic (SAC) agent generates a continuous steering differential-torque demand from the steering input and measurable vehicle states. A subsequent torque-coordination and allocation module combines the two demands subject to the speed-dependent torque limits of the bilateral motors and reconstructs mutually compatible longitudinal-speed and yaw-rate references through the nominal model. In the lower layer, Tube model predictive control tracks the reconstructed references and corrects the bilateral torques under bounded disturbances and actuator constraints. Hardware-in-the-loop tests conducted under the prescribed operating and disturbance conditions indicate that the complete architecture realizes the commanded steering response with smooth torque outputs and provides a favorable balance among state-tracking accuracy, torque smoothness, and disturbance attenuation.