Abstract: Existing research on trajectory tracking control of inspection robots mainly suffers from the following problems: ① insufficient synchronization control accuracy of dual motors under asymmetric load disturbances. ② difficulty of a single control structure in balancing predictive optimization and dynamic disturbance rejection. ③ poor adaptability and robustness of control algorithms under complex and variable road conditions (e.g., varying slopes and surface states). To address these problems, a hierarchical double-closed-loop trajectory tracking control method based on Model Predictive Control (MPC) and Fuzzy Adaptive PID (FAPID) algorithms, namely MPC-FAPID, was proposed. Based on the kinematic model of a four-wheeled differential inspection robot, corresponding constraints were imposed on the control variables and control increments, and an MPC-based trajectory tracking controller was designed. To solve the problem of uncoordinated control caused by wheel speed disturbances in four-wheeled differential robots, the FAPID algorithm was introduced to reduce motor speed errors. The simulation results showed that the FAPID algorithm effectively reduced synchronization error, and its accuracy and robustness were superior to those of PID control and whale-optimized PID control. To overcome the limitation that a single-layer control structure could not balance predictive capability and anti-disturbance performance, a hierarchical double-closed-loop controller based on MPC-FAPID was designed: the outer loop MPC compensated for trajectory tracking errors and handled multiple constraints, while the inner loop FAPID suppressed load disturbance effects. Simulation results indicated that under straight-line motion on a gentle slope, the MPC-FAPID achieved an adjustment time of 0.87 s, which enabled faster convergence of robot pose to the reference trajectory compared with MPC-PID and MPC-whale-optimized PID. Under continuous turning conditions, compared with MPC-PID and MPC-whale-optimized PID, the MPC-FAPID better captured the variations in the reference trajectory, with maximum lateral, longitudinal, and heading angle errors of −0.051 m, 0.000 47 m, and 0.040 8 rad, respectively. Experimental results showed that, compared with MPC-PID, MPC-FAPID reduced the maximum lateral error by 88.24% and the maximum longitudinal error by 87.76% in the multi-target point trajectory tracking test.