A method, apparatus and device for training a model

By using multi-model interaction and differential adjustments, diverse trajectory generation models are trained, solving the adaptability problem of existing trajectory generation models in complex traffic scenarios and improving the decision-making and safety performance of autonomous driving systems.

CN122196533APending Publication Date: 2026-06-12CHONGQING CHANGAN AUTOMOBILE CO LTD
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHONGQING CHANGAN AUTOMOBILE CO LTD
Filing Date
2026-03-03
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

In complex traffic scenarios, existing technologies struggle to adapt trajectory generation models to real-world interaction needs, impacting the decision-making performance and safety of autonomous driving systems and lacking diverse and effective control mechanisms.

⚗Method used

By using multi-model interaction and parameter tuning based on the degree of difference, multiple diverse second trajectory generation models are trained to enhance their diversity. These models are then used to train the initial first trajectory generation model, improving its robustness and generalization ability in complex traffic scenarios.

🎯Benefits of technology

It enhances the diversity and robustness of trajectory generation models, improving the decision-making ability and safety of autonomous driving systems in complex traffic environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

The application provides a method, device and equipment for training a model, the method comprising: based on a plurality of second trajectory generation models last time and an initial first trajectory generation model, respectively controlling a background vehicle and a target vehicle to interact in a simulation driving scene, and generating first trajectory data; based on the first trajectory data, determining first entropy values corresponding to the plurality of second trajectory generation models last time; in the case that the first entropy values do not satisfy a threshold value, adjusting parameters of the plurality of second trajectory generation models last time to obtain a plurality of second trajectory generation models this time; based on the plurality of second trajectory generation models this time, continuing to train a plurality of second trajectory generation models next time until a termination condition is satisfied, and obtaining a plurality of second trajectory generation models last time; and using the plurality of second trajectory generation models last time to train the initial first trajectory generation model, and obtaining a third trajectory generation model for deployment on a first vehicle.
Need to check novelty before this filing date? Find Prior Art