Reinforcement Learning-based Intelligent Decision-making Methods For Unmanned Vehicles

In recent years, with the vehicle’s world-wide popularity, as it brings convenience to people’s lives, frequent traffic accidents also will become an urgent problem, driver assistance systems research is exigent. As the super accidents always take place on highway, the driver assistance systems based on highway environment is significant. For the complexity of the vehicle and environment model and we don’t have many experience of environment, the autonomous vehicle which can obtain right decision based on environment become a hotspot in research fields. In this paper, we propose a method of autonomous decision-making based on reinforcement learning, we also analyze the multi-objective problem of driving and optimize lateral tracking control.The main contributions and innovations are as follows:1. By using the model of an HQ3 autonomous vehicle, we build a simulation platform based on the highway environmental model. Through the simulation platform, we can real-time simulate highway traffic environment. By using the update of the cars in the environment and the uncertain complication, we can improve the system’s suitability. The platform is the basis for proving the decision-making system and follow studies.2. Considering the actual condition, we propose a method of autonomous decision-making based on reinforcement learning. Our research includes the system based on Q and the system based on Least-squares Policy Iteration(LSPI). With the simulation on the simulate platform, we validate the efficiency of the decision-making system.3. For the multi-objective problem of the vehicle driving, we propose a method of autonomous decision-making based on multi-objective reinforcement learning. By the simulation on the simulation platform, we prove the efficiency of the system. Without relearning process, we can get different decision system by easily changing the decision- making part of the system. It can greatly enhance the practicality of the decision-making system.4. Combined with the model of HQ3 autonomous vehicle, the paper design a lateral tracking arithmetic based on LSPI-PD, in order to improve the lateral tracking performance. Based on the optimization of lateral tracking, the paper research the decision-making method combined with the optimization of control system. With the simulation on the simulate platform, we validate the efficiency of the decision-making method combined with the optimization of the lateral tracking control system.