Risk aggregation method based on improved conditional value-at-risk in cross-country scene
By using an improved conditional value at risk (VaR) method and sparse Gaussian process regression, a comprehensive access risk map for off-road scenarios is constructed, which solves the problems of fragmentation and uncertainty in risk assessment in off-road autonomous driving and achieves high-precision and robust risk assessment and path planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INST OF TECH
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies for assessing the risks of off-road terrain traversal in autonomous driving suffer from fragmented assessment standards and a lack of systematic integration. They are unable to accurately capture low-probability, high-loss extreme risk events and do not fully consider the uncertainty of perception data, leading to a decline in the reliability of risk assessment.
An improved conditional value at risk approach is adopted. By establishing a multi-level coordinate system transformation relationship and combining it with the sparse Gaussian process regression method, a 2.5D raster elevation map is constructed. Risk indicators such as flatness, slope and terrain difference are integrated and risk weights are optimized to generate a comprehensive access risk map.
It significantly improves the modeling accuracy and robustness of traffic risk assessment in off-road environments, enhances the physical rationality and local adaptability of risk assessment, and improves the ability to identify low-probability, high-loss terrain, providing a reliable foundation for real-time autonomous path planning.
Smart Images

Figure CN122087690A_ABST