High-risk autonomous driving navigation strategy based on heterogeneous sensor collaboration and motion state constraints

Facing the challenges of heavy-duty vehicles in complex industrial areas, such as strong magnetic vibrations and instability under full-load liquid conditions, there is an urgent need to overcome the bottlenecks of cross-modal perception distortion and the disconnect between purely mathematical trajectories and physical execution. To address the challenges of autonomous driving in high-risk environments, this study proposes an autonomous driving navigation strategy based on heterogeneous sensor collaboration and motion state constraints. This strategy cleans multi-source distortion data through Extended Kalman Filter collaborative solution, reconstructs the visual network integrated with the spatial attention mechanism to enhance the capture of hidden hazard source features. It also deeply couples the anti-sway extreme value of the heavy-duty chassis and the non-integrity constraints into the optimization cost matrix. The results showed that the absolute trajectory error of this strategy under extreme intensity noise was only 0.38m, effectively locking the robust spatial benchmark. The average accuracy of target detection in heavy occlusion scenes reached 88.2%. Local dynamic avoidance actively smoothly converged the peak deceleration of heavy truck emergency braking to -2.31m/s², effectively avoiding internal liquid impact instability. The research provides a feasible risk-avoidance paradigm for the unmanned evolution of heavy-duty machinery in high-risk pipeline corridors, and accurately opens up a theoretical closed loop of deep collaboration between cross-modal perception calculation and underlying mechanical execution.