A method and system for automatic driving decision-making control based on layered reinforcement learning

Abstract

The invention discloses an automatic driving decision control method and system based on layered reinforcement learning, including an action layer Agent and a decision layer Agent. The action layer Agent is a policy gradient-based reinforcement learning algorithm that deals with continuous behavior, and trains the correct action network by interacting with road environment information; the decision-making layer Agent chooses a value-based reinforcement learning algorithm that handles discrete behavior, and uses traffic flow and traffic status information Interactively train the correct decision network. By first training the action layer Agent, the decision layer Agent is trained on the basis of the already trained action layer Agent, so that the two training tasks are clear and the learning efficiency is improved. The invention avoids sudden acceleration and deceleration, and improves the comfort of the whole vehicle.

Description

technical field [0001] The invention belongs to the field of automatic driving of intelligent vehicles, and relates to an automatic driving decision-making control method system based on layered reinforcement learning. Background technique [0002] Reinforcement learning is a rapidly developing machine learning method that emphasizes selecting an action based on the current state of the environment so that the action can maximize the expected reward. It is a trial-and-error learning method. During the learning process, through the stimulation of rewards, one can gradually make actions that maximize the expected rewards. Among them, the model-free reinforcement learning method has attracted much attention because it does not require modeling and has good progressive performance. The DDPG and DQN algorithms are two different model-free reinforcement learning methods. DDPG is a policy gradient-based reinforcement learning algorithm for continuous behavior, and DQN is a value-b...

AI Technical Summary

Problems solved by technology

[0003] Traditional automatic driving control methods are difficult to make correct decisions and actions taking into account the complex environment, and have gradually failed to meet the decision-making and control requirements of automatic driving.

Images