Automatic driving controller and training method based on variational auto-encoder and reinforcement learning

Abstract

The invention discloses an automatic driving controller and a training method based on a variational auto-encoder and reinforcement learning. The variational auto-encoder is used for extracting surrounding traffic environment information, the encoder adopts a method of a convolutional neural network and a recurrent neural network, information of multiple sensors and historical environment information are effectively extracted, and the information loss is avoided. The reinforcement learning network uses a potential variable extracted by dimension reduction of a variational auto-encoder as a state quantity for training, and the problem that the state space of a reinforcement learning part is too large is solved. The additional reward constructed by using the loss function of the variational auto-encoder accelerates the exploration of the intelligent agent to the unfamiliar state space, and improves the exploration rate and learning rate of reinforcement learning.

Description

technical field [0001] The invention belongs to the technical field of automatic driving vehicles, and in particular relates to an automatic driving controller and a training method based on variational autoencoders and reinforcement learning. Background technique [0002] As a main research content in the field of intelligent traffic control, intelligent vehicle integrates a variety of modern electronic information technologies. With the current society's increasing demand for intelligent and safe modern vehicles, intelligent driving has become a hot issue and technological frontier in the field of transportation in various countries in the world. As a rapidly developing machine learning algorithm, reinforcement learning has been applied to the field of intelligent driving by more and more experts and scholars. [0003] Reinforcement learning is a rapidly developing machine learning method that emphasizes selecting an action based on the current state of the environment so...

AI Technical Summary

Problems solved by technology

[0004] However, due to the huge amount of environmental information and the complexity of the traffic scene in the process of automatic driving, the direct application of the method of reinforcement learning to the field of automatic driving has a large state quantity space and the learning efficiency caused by the low rate of exploration of the strange state space by the agent. A series of low-level questions

Images