Unmanned vehicle intelligent migration method and system based on developmental clustering

Abstract

The invention discloses an unmanned vehicle intelligent migration method and system based on developmental clustering. The method comprises the steps: S1, obtaining vehicle input sensing information in a historical known environment in a learning stage, outputting operation information, and developing an initial clustering result in an incremental form, obtaining a corresponding relationship between each type of input perception information and output operation information, and constructing and forming an input and output association network; s2, when the input perception information needs to be migrated to an unknown environment, collecting newly added input perception information in the unknown environment, clustering the newly added input perception information with various input perception information in the input and output association network, and outputting output operation information corresponding to the input perception information similar to the newly added input perception information according to a clustering result, and migrating to the newly added input perception information. Intelligent and quick migration of the unmanned vehicle can be realized so that the unmanned vehicle is enabled to have efficient migration capability.

Description

technical field [0001] The present invention relates to the technical field of migration of unmanned vehicles, in particular to a method and system for intelligent migration of unmanned vehicles based on developmental clustering. Background technique [0002] Active safety technology for the normal driving process of unmanned vehicles to prevent traffic accidents has become the focus of current research on unmanned vehicle technology. The main difficulty currently faced by unmanned vehicles is how to deal with various situations in an unknown environment and make reasonable judgments in a timely manner. Therefore, it is very necessary for unmanned vehicles to have real-time perception of the dynamic changes of the main participants such as vehicles, traffic signs, and pedestrians in the traffic environment. Control decision-making strategies, directly intervene in vehicle control when necessary, prevent traffic accidents, and protect the lives and property safety of drivers...

AI Technical Summary

Problems solved by technology

If the prior parameter setting is unreasonable, it will lead to learning failure, and the algorithm needs to be adjusted and restarted

(2) The ability to subdivide the data is insufficient: in the sample space with a small amount of data, the difference of samples in the cluster will be enlarged, and the algorithm should actually tend to distinguish samples; The difference between samples is reduced in a larger sample space, and the algorithm should actually be inclined to the fusion between samples

However, the results obtained by the batch clustering algorithm are fixed, and the subdivision ability will not change with the increase of samples, and it cannot make adaptive adjustments as the amount of data changes.

(3) Algorithms fail when small samples are clustered: too few samples cause most algorithms to fail, that is, developmental learning starting from 0 samples cannot be realized

(4) The phenomenon of forgetting is difficult to overcome: if new samples are added to the learned data set, it may destroy the learned results, especially for clustering methods that need to obtain clustering structure parameters through preprocessing

Therefore, using the traditional deep model-based migration method in the unmanned vehicle, the migration of the unknown environment is independent of the learning process of the known environment, and it is impossible to use the learning results of the known environment to achieve a reasonable judgment on the unknown environment, nor can it Adjust the learning results of the known environment according to the information of the unknown environment, resulting in poor transfer performance

Images