To enhance the motion stability and environmental adaptability of inspection robots in complex terrains of oil and gas fields, and to improve their obstacle-crossing performance, this research conducted an in-depth analysis of a three-layer collaborative control architecture and its application in an inspection robot system. The D-H parameter method was employed to construct a joint coordinate system chain and generate a displacement control matrix for high-precision positioning. The integration of fifth-order polynomial trajectories with model predictive control rolling optimization was used to suppress roll and pitch vibrations in real time. A grid elevation perception model was fused to optimize the landing point sequence for minimizing path slope, ultimately forming a perception-control closed-loop system that significantly enhanced the robot's stability and adaptability on steep slopes and gullies. Experimental results demonstrated that the proposed control system effectively improved steering instability and trajectory deviation, with body attitude fluctuation not exceeding 0.02 rad. During dynamic stability tests, the pitch angle swing remained near 0 rad, and the climbing error rate was below 2.5%, indicating excellent control performance and environmental adaptability. The three-layer collaborative control system developed in this study provides reliable technical support for safe robot inspections in high-risk oil and gas field environments and significantly enhances the robot's obstacle-crossing capability in complex terrains.