Method and apparatus for training a control strategy model for generating an automatic driving strategy

Abstract

This application relates to the field of artificial intelligence and unmanned driving, and provides a method and device for training a control strategy model for generating an automatic driving strategy. First, the latent feature space and the hidden feature space defined on the latent feature space are determined by using low-dimensional training data. The policy function, which then targets the latent feature space, supervises the training of an encoder that maps high-dimensional training data to the latent feature space, and then applies the encoder and the above policy function to the real traffic environment, i.e., input from real High-dimensional data obtained in the traffic environment, so that high-dimensional data can be used to directly obtain available automatic driving strategies.

Description

technical field [0001] The present application relates to the field of automatic driving, and in particular to a method and device for training a control strategy model used to generate an automatic driving strategy. Background technique [0002] Autonomous driving is a technology in which computer systems replace humans to drive motor vehicles, including functional modules such as environmental perception, location positioning, path planning, decision-making control, and power systems. Among them, the ways to realize the environment perception function include the following two: realize the environment perception function through high-precision and low-dimensional sensors such as laser radar and millimeter-wave radar, and realize the environment perception through high-dimensional and low-precision sensors such as monocular / multi-eye high-definition cameras Features. [0003] In general, high-precision and low-dimensional sensors such as lidar are expensive and their accur...

AI Technical Summary

Problems solved by technology

[0004] However, due to the large amount of information contained in high-dimensional data, usually, high-dimensional data also contains some redundant information. Therefore, it is difficult to directly obtain usable automatic driving strategies by processing high-dimensional data through artificial neural networks.

Images