A data–model collaborative fault-tolerant control method for four-in-wheel-motor unmanned ground vehicles under wheel-corner actuator module faults

Four in-wheel motor drive systems allow fully independent control of wheel torques, thereby improving the maneuverability of unmanned ground vehicles (UGVs). However, the introduction of multi-sensor architectures and distributed actuators increases system complexity, which raises the probability of motor failures and poses challenges to vehicle stability. To address this issue, this paper proposes a data–model collaborative fault-tolerant control method for four-in-wheel-motor UGVs under wheel-corner actuator module fault conditions. A Bayesian Transformer-based adaptive compensation mechanism is developed and embedded into the MPC framework. The residual sequence generated from the discrepancy between the nominal vehicle model and measured wheel responses is combined with vehicle states, control inputs, and fault flags as the input of the Bayesian Transformer. The network estimates the mean and uncertainty of future model residuals, and the predicted residual mean is introduced into the MPC prediction model as an online compensation term to correct fault-induced model mismatch. Based on the identified fault states, the controller adaptively adjusts wheel torque distribution to maintain stable vehicle motion under wheel-corner actuator module fault conditions. Simulation results demonstrate that the proposed method enhances fault-tolerant control performance and trajectory tracking accuracy under both single-module and dual-module wheel-corner actuator faults. Compared with conventional MPC, the proposed method significantly reduces the mean lateral and longitudinal tracking errors under both single-module and dual-module fault conditions. These results indicate that the proposed data–model collaborative control framework provides an effective fault-tolerant solution for UGV operation in complex and degraded environments, and offers methodological support for UGV applications in high-risk autonomous tasks.