基于多源数据融合的无人系统轨迹预测可信评估方法与装置
By fusing multi-source data and using an improved gradient attack algorithm, combined with the physical characteristics of trajectory data and an attention-guided loss function, the reliability of trajectory prediction models for unmanned systems is evaluated. This solves the problems of uninterpretability and safety hazards of deep learning models in unmanned systems, and enables reasonable decision-making in complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INST OF SPACECRAFT SYST ENG
- Filing Date
- 2025-07-03
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning trajectory prediction models have uninterpretability and safety risks in unmanned systems, especially when perturbations are added to the trajectory prediction, causing the predicted trajectory to deviate significantly, making it difficult to assess the reliability of the model.
A multi-source data fusion-based approach is adopted. The model is trained by constructing a trajectory prediction dataset to generate an adversarial example set. An improved gradient attack algorithm and an attention-guided loss function are used to screen out the consistent adversarial examples with the largest average displacement error for credibility assessment.
Effective detection and evaluation of the reliability of trajectory prediction models for unmanned systems ensures that they make reasonable decisions in complex environments, thereby improving the robustness of the models and the relevance of the evaluation.
Smart Images
Figure CN120849889B_ABST