• Deep RL model optimizes LiDAR resolution for autonomous robot exploration. • Improved mapping accuracy in unknown and dynamic terrains. • Adaptive learning enhances exploration efficiency and decision-making. • Results support safer, data-efficient autonomous navigation systems. This research proposes a multi-model combinatorial congestion mitigation approach to solve the traffic safety problem and speed optimisation problem of autonomous vehicles. The proposed model framework consists of three parts: a road modelling module based on meta cellular automata, a traffic flow speed prediction module consisting of an optimised long and short-term memory algorithm using a sparrow search algorithm, and a following distance prediction module that determines the optimal safe following distance using an adaptive cruise control algorithm. In the simulation, this intelligent traffic model for autonomous cars can be applied to various complex traffic scenarios based on the design of real traffic intersections, which improves traffic efficiency while reducing the collision risk of autonomous vehicles. After setting up a control group experiment to verify, compared with the traditional optimisation algorithm, the algorithm model designed in this study significantly improves the prediction ability of the autonomous vehicle when subjected to traffic pressure. In the one-hour simulation experiment process, the vehicle’s average speed was increased by at least 3.89%. At the same time, the judgement of the safety distance of the following car was also made ahead of time, which made the vehicle drive more smoothly. When dealing with traffic congestion caused by accidents on the road, the average queue length of vehicles was reduced by 92.86%, and the maximum queue length was decreased by 78.57%.
Yufei et al. (Sun,) studied this question.