Enhancing Obstacle Detection and Control in Autonomous Robotic Vehicles Through Edge Computing Integration

Especially with the continuous maturity of the autonomous driving technology, the construction of safety evaluation model that can effectively solve accurate obstacle detection and control while ensuring real-time performance has become an urgent problem in the field of intelligent transportation. Based on the existing multi-task convolutional neural network combined with a reinforcement learning model, this paper presents a more complex and strongly integrated edge computing-driven intelligent perception and control framework. Iniyially, advanced ResNet network is employed to extract the features of multi-source sensor input, and the dynamic attention mechanism is adopted at this stage to filter out important information of obstacles. Then, through the combination of a hierarchical policy gradient-based reinforcement learning algorithm, the adaptive adjustment of vehicle speed and steering was completed, and the interactive iteration between the local decision-making layer and the global scheduling layer was carried out. Another is the collaborative detection part, where distributed data preprocessing and model inference modules are deployed on the edge side at the same time, and high-dimensional images and lidar point clouds are fused in real time in this area through lightweight parallel computing strategy. Experimental results indicate that the model realizes high-precision obstacle detection and stable vehicle control in complex road scenarios, and the overall performance exceeds that of existing mainstream methods.