Search papers, labs, and topics across Lattice.
The paper introduces three novel effort-based criticality metrics鈥擣alse Speed Reduction (FSR), Maximum Deceleration Rate (MDR), and Lateral Evasion Acceleration (LEA)鈥攖o better evaluate 3D perception errors in autonomous driving scenarios. These metrics quantify the effort required to mitigate the consequences of false-positive and false-negative perception errors, addressing limitations of existing time-based metrics. Experiments on nuScenes and Argoverse~2 datasets demonstrate that these metrics effectively identify safety-critical perception failures that are missed by traditional approaches.
Current autonomous driving metrics miss the most dangerous perception errors: effort-based metrics reveal that up to 93% of errors are non-critical and identify the failures that truly matter for safety.
Criticality metrics such as time-to-collision (TTC) quantify collision urgency but conflate the consequences of false-positive (FP) and false-negative (FN) perception errors. We propose two novel effort-based metrics: False Speed Reduction (FSR), the cumulative velocity loss from persistent phantom detections, and Maximum Deceleration Rate (MDR), the peak braking demand from missed objects under a constant-acceleration model. These longitudinal metrics are complemented by Lateral Evasion Acceleration (LEA), adapted from prior lateral evasion kinematics and coupled with reachability-based collision timing to quantify the minimum steering effort to avoid a predicted collision. A reachability-based ellipsoidal collision filter ensures only dynamically plausible threats are scored, with frame-level matching and track-level aggregation. Evaluation of different perception pipelines on nuScenes and Argoverse~2 shows that 65-93% of errors are non-critical, and Spearman correlation analysis confirms that all three metrics capture safety-relevant information inaccessible to established time-based, deceleration-based, or normalized criticality measures, enabling targeted mining of the most critical perception failures.
