Physics-Based Plausibility Assessment of AI-Predicted Vehicle Parameters for Automotive Radar in Pre-Crash Scenarios
Physics-Based Plausibility Assessment of AI-Predicted Vehicle Parameters for Automotive Radar in Pre-Crash Scenarios
This research presents a framework for validating AI-derived vehicle hypotheses in automotive radar perception by generating expected radar signatures from AI-estimated parameters using a physics-based reference model. The approach compares these expected signatures with observed radar point clouds using a Wasserstein-distance-based consistency metric derived from spatial distribution, local detection count, and reflection intensity. This allows identification and correction of physically implausible object hypotheses independent of the neural network architecture, without requiring model retraining or architectural modifications. The method is specifically applied to vehicle heading estimation in pre-crash scenarios.
The paper introduces a physics-based consistency module that validates AI-derived vehicle hypotheses in automotive radar by comparing expected signatures against observed point clouds using Wasserstein distance, aiming to correct physically implausible estimates without retraining the underlying neural network.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org