Dynamic Obstacle Avoidance Path Decision-Making Method for Vehicle Navigation based on Vehicle-Infrastructure Cooperative Learning and Edge Computing

Abstract

In dynamic traffic environments, existing navigation methods find it difficult to achieve low latency and high reliability dynamic obstacle avoidance path decision-making in complex scenarios due to the limited perception range of individual vehicles and on-board computing capabilities. To address this problem, this study suggests a dynamic obstacle avoidance path decision approach to vehicle navigation, which is based on the Vehicle-Infrastructure Cooperation (VIC) and Edge Computing (EC). Through VIC, the vehicle status and roadside perception information are obtained and a unified spatio-temporal traffic map model is constructed at the edge nodes. This article maps the spatial position and motion trend of dynamic obstacles to a continuously updated environmental cost field and introduces Model Predictive Control (MPC) for rolling optimisation based on this. Constraints of obstacle avoidance, dynamics of the vehicle, as well as the navigation objectives, are all integrated in the finite time domain optimisation problem. The edge side continuously outputs the optimal local path and issues it to the vehicle for execution in real time, forming a closed-loop updated path decision-making process that changes with the environment. The experimental results show that the proposed method achieves a path planning delay of 20.6-24.1 ms in typical urban scenarios containing 10 to 14 dynamic obstacles, with a dynamic obstacle avoidance success rate of over 91.5%. At the same time, the path tracking error is controlled within the range of 0.22-0.35 m, verifying the effectiveness of the method in real-time, safety and stability. This article provides a feasible technical path for intelligent navigation and dynamic decision-making in VIC environment.