Automatic driving overtaking decision-making method based on reinforcement learning under opposite double lanes

Abstract

The invention discloses an automatic driving overtaking decision-making method based on reinforcement learning under opposite double lanes. The method comprises the following steps of collecting the traffic state of an automatic driving vehicle through a sensor; inputting the collected traffic state into a trained decision model; enabling the decision model to select a corresponding driving actioninstruction from the action space according to the input information and output the driving action instruction, and driving the vehicle automatically to form a new traffic state after the driving action; calculating a reward value of the driving action through a reward function, and storing the original traffic state, the driving action, the reward value and the new traffic state into an experience playback pool as transfer samples; calculating a loss function value of the decision model, and optimizing parameters of the decision model according to the transfer sample and the loss function value; and repeating the steps until the automatic driving is finished. The safety and comfort of the overtaking decision-making process of the automatic driving vehicle are ensured, and the humanization and robustness of decision-making are improved through a reinforcement learning decision-making method.

Description

technical field [0001] The invention relates to the technical field of automatic driving, in particular to an overtaking decision method for automatic driving based on reinforcement learning in opposite dual lanes. Background technique [0002] The self-driving vehicle system generally adopts a layered structure, which consists of three modules: environment perception, decision planning, and vehicle control. The environment perception module detects obstacle information in the traffic environment and generates an environment map to determine the drivable area. The decision planning module is equivalent to the brain of the vehicle. The decision-making module is responsible for decomposing and generating various discrete driving tasks of self-driving vehicles, such as overtaking, cruising, collision avoidance, acceleration and deceleration, and other macro-action commands. Once the current driving task is determined, the continuous vehicle trajectory is planned on the basis ...

AI Technical Summary

Problems solved by technology

However, this method also has obvious limitations: it cannot deal with uncertain scenarios

Therefore, the overtaking decision in this traffic environment is more complicated than the simple expressway environment. The traditional state machine method will not be able to deal with the uncertain factors in the environment. A transfer error occurred

Images