Automatic driving decision-making method and automatic driving device of vehicle

Abstract

The invention discloses an automatic driving decision-making method and an automatic driving device of a vehicle. Compared with a common feedforward control decision-making method, the automatic driving decision-making method and the automatic driving device of the vehicle can enable a driving system to be trained under a sufficient data condition based on a deep reinforcement learning method, andcan quickly and accurately make a decision under a higher-speed and complex road surface environment after training. Different from an existing reinforcement learning method in which huge loads are generated due to centralized processing of trajectory generation and tracking tasks, and a system is not converged due to execution of the current state of a vehicle, the method has the advantages thatthe feedforward error amount of a target vehicle is directly used as the state of the target vehicle, an immediate return obtained by executing the action of optimal strategy mapping in the current state is set as an opposite number of the current feedforward error amount so as to carry out deep reinforcement learning; trajectory planning and action planning are decoupled, on the premise that trajectory planning is completed, the tracking problem after trajectory planning is solved in a centralized mode, an optimal strategy is obtained, a strategy network model is constructed, and convergenceand self-adaptation of the model can be achieved.

Description

technical field [0001] The invention relates to the technical field of automatic driving, in particular to an automatic driving decision-making method and an automatic driving device for vehicles. Background technique [0002] Autonomous driving is an inevitable trend of future vehicle development and an effective way to avoid human driving errors and improve traffic efficiency. In order to realize the safe driving of unmanned vehicles on real roads, not only the vehicle body perception end needs to accurately identify and track objects, but also the decision-making control end needs to be able to make reasonable actions according to the current road conditions, including behavior prediction, avoidance, etc. Obstacles, trajectory planning, action planning and executive agencies, etc. [0003] Because the driving scene is relatively complex, the driving behavior is usually complex and changeable, and traffic and road emergencies also occur from time to time, it is difficult ...

AI Technical Summary

Problems solved by technology

However, the existing technology performs reinforcement learning for the current state of the vehicle such as heading angle and longitudinal speed. There is no preview point and no model, which often leads to the system not converging or adaptive, and the obtained strategy is not accurate.

Images