Method for predicting action of movable platform and method for training action prediction model

By constructing a unified potential space vision-language-action model in the autonomous driving system, and combining the characteristics of long-term planning and short-term control, the problem of inaccurate action prediction is solved, achieving more accurate and timely action prediction, and improving the foresight and stability of the autonomous driving system.

CN122336705APending Publication Date: 2026-07-03北京极佳视界科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
北京极佳视界科技有限公司
Filing Date
2026-04-01
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies for autonomous driving suffer from inaccurate motion prediction, particularly due to the failure to effectively decouple long-term planning from short-term control. This leads to error accumulation and insufficient response to sudden environmental changes, affecting the safety decisions and traffic efficiency of autonomous driving systems.

Method used

By acquiring multimodal dynamic information from mobile platforms, visual latent features and action latent features in a unified latent space are constructed using a vision-language-action pre-trained model. Action prediction is then performed by combining the prediction latent variable features from long-term planning with short-term real-time features. A hierarchical collaborative mechanism is adopted for action prediction, which includes the combined use of multiple encoders and pre-set models.

Benefits of technology

It achieves more accurate and timely action prediction, improves foresight, real-time performance and stability, solves the decoupling problem between long-sequence planning and short-sequence control, and enhances the decision-making and environmental response capabilities of autonomous driving systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122336705A_ABST
    Figure CN122336705A_ABST
Patent Text Reader

Abstract

The embodiments of the present disclosure disclose a movable platform action prediction method and a training method of an action prediction model, wherein the movable platform action prediction method comprises: acquiring multi-modal dynamic related information corresponding to the movable platform at a current time; determining visual latent features and action latent features in a unified latent space based on the dynamic related information, realizing short-time sequence real-time response; performing prediction processing on at least one of the multi-modal dynamic related information to obtain prediction latent variable features at future k time points relative to the current time, obtaining k prediction latent variable features, and realizing long-time sequence dynamic perception planning; and performing action prediction based on the visual latent features, the action latent features and the k prediction latent variable features to obtain an action prediction result at a next time point relative to the current time. The embodiments of the present disclosure improve the foresight, real-time and stability of action prediction.
Need to check novelty before this filing date? Find Prior Art