Automatic driving key scene generation method based on reinforcement learning

Abstract

The invention discloses an automatic driving key scene generation method based on reinforcement learning, and the method comprises: 1), selecting a road scene from a map library, setting a driving route of a main vehicle in a simulation system, and building a probability model for each dynamic environment element; 2) controlling the main vehicle by the simulation system to start to execute a simulation task, and, based on a reinforcement learning technology, training the probability model of each dynamic element in the selected road scene to obtain an optimal parameter of each probability model for the selected road scene, and storing the optimal parameter in a test case library; 3) circulating the steps 1-2) to obtain optimal parameters of each probability model for each road scene in the map library; 4) obtaining a plurality of road scenes from the map library, combining the road scenes to obtain a test map, and selecting dynamic elements required in a simulation environment; and 5) importing the probability model of each dynamic element contained in the test map and the corresponding optimal parameter from the test case library, and generating a key scene test case.

Description

technical field [0001] The invention relates to a method for generating key scenes of automatic driving based on reinforcement learning, which belongs to the technical field of computer software. Background technique [0002] Today, the performance of most perception and prediction algorithms is very sensitive to the imbalance of training data (also known as the long-tail problem), and rare events are usually difficult to collect and easily ignored in the huge data stream, which greatly challenges Real-world applications of robots, especially in safety-critical domains such as autonomous driving. [0003] In the autonomous driving industry, simulations are often used to reproduce key scenarios collected during test drives. The prior art proposes an alternative method called worst-case evaluation to search for worst-case evaluated controllers in the vehicle domain. While some of the situations unearthed through worst-case assessments may be useful, some extremely risky situ...

AI Technical Summary

Problems solved by technology

[0005] The prior art lacks the training of dynamic environment elements in the simulation environment of automatic driving, and lacks the problem of how to deploy the dynamic environment elements to generate key safety scenarios for automatic driving. The purpose of the present invention is to provide a key scene of automatic driving based on reinforcement learning generation method

Images