Smart algorithms such as path planning are faced with challenges such as traffic congestion, dynamic environmental obstacles change, high computational complexity and multi-agent collaboration. To solve these problems, a method of multi-agent path planning and angle tracking strategy generation system is proposed, which integrates hybrid algorithm, dual-stream network, dynamic weight allocation and hierarchical learning, and cloud-edge collaborative training. The results show that under the full model configuration, the path success rate is as high as 98.73%, and the conflict rate is only 1.25%. In the high obstacle density environment, the trajectory jitter rate is stable at about 0.3%, and the entropy of angular deviation is about 0.8 Bits. In the communication delay robustness test, the synchronization error at 200ms delay was only half that of the comparison scheme, and the overlap rate was also about 30% lower. The proposed method shows higher stability and accuracy in complex environments, which significantly improves the performance of multi-agent path planning and angle tracking, and has stronger robustness of communication delay.