| During the driving process,it is often the case that the self-driving car cannot perceive the specific traffic state ahead due to the obstruction of objects in the road environment,and thus a traffic accident occurs.Therefore,in order to avoid decision-making errors or delays caused by the occlusion of the preceding vehicle and other blind areas,autonomous vehicles need to use the cognitive information obtained by the preceding vehicle or the roadside to assist decision-making when making driving decisions.The existing end-to-end algorithm for multi-vehicle collaborative decision-making mainly relies on a large amount of labeled data for joint training,but it does not fully explore the relationship between multi-source data,which makes the efficiency and interpretability of the end-to-end decision algorithm not high.In response to this problem,the paper proposes a multi-vehicle collaborative trans-horizon cognition method based on the attention mechanism,which is mainly composed of two parts:(1)Taking into account the effectiveness of cognitive information transmission in the car-shop and the importance of different information sources in different scenarios,a single-data multi-vehicle collaborative decision generation based on a scene-adaptive cognitive attention mechanism is proposed.In the algorithm,the deep neural network outputs the cognitive intermediate results as the cognitive information to be exchanged,and realizes the joint decision-making of influence cognition based on self-cognition information and beyond-the-horizon cognitive information.(2)Aiming at the problem of cognitive differences caused by the different environments of multi-sensing sources,a multi-vehicle collaborative data adaptive filtering algorithm is proposed,so that the neural network can select the adaptive data type according to different light intensity,and reduce the performance of the sensor in a specific scene.Data interference caused by poor performance.In addition,because the existing data set contains only single self-driving cars,use the Carla city driving simulator to simulate the collaborative work scene between self-driving vehicles and record all the sensor data and the research on the subject in the course of the vehicle.Finally,the paper carried out simulation and comparative experiments on the proposed method in the Carla city driving simulator.The experimental results showed the effectiveness of the multi-vehicle collaborative beyond-the-horizon cognitive method based on the attention mechanism proposed in this paper. |
