A multi-modal trajectory planning method based on hierarchical query optimization and anchor iteration

CN122384802APending Publication Date: 2026-07-14SOUTHEAST UNIV
View PDF 0 Cites -1 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTHEAST UNIV
Filing Date
2026-04-14
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing autonomous driving trajectory planning methods struggle to effectively model the spatiotemporal relationships between dynamic obstacles, static obstacles, and map elements in complex traffic scenarios. They often employ single-modal trajectory output, resulting in monotonous driving behavior, an inability to cope with ambiguity and uncertainty, and the generated trajectories are prone to deviating from road topology constraints.

Method used

A multimodal trajectory planning method based on hierarchical query optimization and anchor point iteration is adopted. A multimodal trajectory planning model is constructed through the Transformer architecture. Dynamic obstacles, static obstacles and map elements are independently encoded and interacted and fused through a self-attention mechanism to generate a unified environmental context representation. An initial query vector is constructed by combining reference line anchors. The query vector is refined layer by layer and the anchor position is iteratively updated. Finally, the multimodal candidate trajectory is decoded and generated.

Benefits of technology

It achieves effective modeling of the spatiotemporal dependencies of various elements in complex traffic scenarios, generates multimodal candidate trajectories with multiple driving strategies, the trajectories are closer to human driving behavior, follow road topology constraints and have micro-adjustment capabilities, thus improving the rationality and safety of the trajectories.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122384802A_ABST
    Figure CN122384802A_ABST
Patent Text Reader

Abstract

The application discloses a kind of multi-modal trajectory planning methods based on layered query optimization and anchor iteration, it is related to the field of automatic driving technology.The application includes receiving the scene intermediate feature data with ego vehicle as center and carrying out preprocessing operation, form structured multi-modal feature input;Multi-modal trajectory planning model is constructed based on the architecture of Transformer, and loss function is constructed to train and optimize the model, obtain the trained multi-modal trajectory planning model.The application effectively models the space-time interaction relationship of each element in complex traffic scene by independently encoding and self-attention fusion of multi-modal features;Through anchor initialization based on reference line key point and learnable mode offset, the parallel generation of multiple driving strategies is realized;Through the layer-by-layer refinement of query vector by multiple optimization layers, combined with dynamic position encoding and the space perception mechanism from coarse to fine, the generated trajectory not only follows the road topology constraint but also has microscopic adjustment capability.
Need to check novelty before this filing date? Find Prior Art