Deep reinforcement learning intelligent vehicle behavior decision-making method based on path planning

Abstract

The invention discloses a deep reinforcement learning intelligent vehicle behavior decision-making method based on path planning, and belongs to the technical field of intelligent vehicle automatic driving, the deep reinforcement learning intelligent vehicle behavior decision-making method based on path planning comprises the steps of modeling a task as a Markov decision-making process, building a deep reinforcement learning algorithm, and making an intelligent vehicle behavior decision-making process. Designing agent input, designing agent output, building a training network structure, performing path planning on a task environment, improving a reward function, and training and testing an agent model; the method has the advantages that complex decisions are processed, the process from a simulation scene to practical application is simplified, the problems of low training speed and difficulty in convergence are solved, and the practical generalization ability of the intelligent agent model is improved.

Description

technical field [0001] The invention relates to the technical field of intelligent vehicle automatic driving, in particular to a path planning-based deep reinforcement learning intelligent vehicle behavior decision-making method. Background technique [0002] In the face of increasingly severe traffic congestion, driving safety and environmental pollution, autonomous driving smart cars have become an inevitable trend in the development of the automotive industry. At the same time, autonomous driving plays a role in promoting industrial prosperity, economic development, technological innovation, and social progress. Major countries have raised it to the level of national strategy. The system architecture of a smart car mainly includes a perception module, a decision-making module, and a control module. The decision-making module, as the "brain" of an unmanned vehicle, is a direct reflection of the intelligence of the autonomous driving system and plays a decisive role in the ...

AI Technical Summary

Problems solved by technology

This optimal strategy is guided by the reward function, and the sparse reward in the learning process will cause the agent to fail to learn a good strategy. Therefore, the design of the reward function is an important part of reinforcement learning, and how to design the reward function reasonably is also important. One of the main research directions of reinforcement learning

Images