Behavior Decision-making Method for Autonomous Vehicle

Behavioral decision-making is a key technology for autonomous driving and determines the safety and stability of the vehicle largely. Current behavioral decision-making solutions, both model-driven and data-driven approaches, is with certain drawbacks. To improve the interpretability and generalization ability of autonomous vehicle behavior decisions, a combined model-driven and data-driven behavior decision method is proposed. Firstly, a CNN-LSTM parallel network model is established to extract the temporal and spatial features in vehicle driving information, then the deep features of driving behavior rules are mined by a GBDT based algorithm. Finally, the extracted driving behavior rules are used to enhance the spatiotemporal characteristics of vehicle driving information and improve the accuracy of self-driving vehicle behavior decisions. The experimental results show that the method can effectively improve the accuracy of self-driving behavior decision results.