Automatic driving model based on hybrid low-rank experts and multi-domain adaptation fine-tuning method

By using a hybrid low-rank expert autonomous driving model, which utilizes a frozen general model base and pluggable low-rank adapter experts to dynamically optimize model decisions, the performance degradation and high cost issues in cross-domain deployment are resolved, achieving flexible and efficient multi-domain adaptation.

CN121523068BActive Publication Date: 2026-04-17TONGJI UNIV
1 Cites 0 Cited by

Patent Information

Application Number
CN202610050836.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-01-15
Publication Date
2026-04-17
Estimated Expiration
2046-01-15

AI Technical Summary

Technical Problem

Existing end-to-end autonomous driving models suffer severe performance degradation when deployed across domains due to domain drift, and traditional methods suffer from high computational and storage costs and catastrophic forgetting issues.

Method used

An autonomous driving model employing hybrid low-rank experts, including a frozen general model base and trainable multi-domain adaptation components, dynamically activates experts to output differential correction signals through pluggable low-rank adapter experts and domain perception routers, thereby optimizing the decision-making of the general model.

Benefits of technology

It enables a single general model to adapt quickly and flexibly to multiple downstream domains, reduces computation and storage costs, avoids catastrophic forgetting, and achieves efficient multi-domain adaptation.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

This application relates to the field of image data processing technology, and particularly to an autonomous driving model based on hybrid low-rank experts and a multi-domain adaptation fine-tuning method. The model includes: a general model base in a frozen state and a multi-domain adaptation component in a trainable state. The general model base includes: an image encoder, a measurement information encoder, and a trajectory planner. The multi-domain adaptation component includes: a hybrid low-rank expert group and a domain-aware router. The hybrid low-rank expert group is used to calculate differential correction signals based on input features. The domain-aware router is used to receive shared image features extracted by the image encoder, output probability vectors, and dynamically activate the corresponding experts in the hybrid low-rank expert group according to the feature distribution of the input data. This application achieves highly efficient "plug-and-play" adaptation of a single general model to multiple downstream domains with extremely low computational and storage costs, perfectly solving the pain points of traditional multi-domain adaptation methods.
Need to check novelty before this filing date? Find Prior Art