Modified-Emergency Index (MEI): A Criticality Metric for Autonomous Driving in Lateral Conflict

TL;DR

The paper addresses the challenge of evaluating autonomous driving safety in urban lateral conflicts by introducing the Modified-Emergency Index ($MEI$), a metric that quantifies evasive effort through the ratio $MEI = rac{InDepth}{TEM}$ where $InDepth$ captures potential intrusion into safety regions and $TEM$ is estimated via TTC$^{2D}$. It extends the concept of Interaction Depth and integrates a safety region with $D_{ ext{safe}}$, using a lateral-conflict dataset derived from Argoverse-2 to validate $MEI$ and compare it against ACT and PET, reporting that $MEI$ more accurately tracks risk and its evolution. The study analyzes 1,548 lateral-conflict events (501 critical, 1,047 potential) and provides percentile-based risk thresholds, demonstrating MEI's superior discrimination of high-risk cases. Open-source code is available at GitHub, enabling practical adoption for urban AV safety assessment. Limitations include a primarily case-study validation and a need for cross-scenario and accident-correlation analyses to further establish generalizability and integrated safety evaluation.

Abstract

Effective, reliable, and efficient evaluation of autonomous driving safety is essential to demonstrate its trustworthiness. Criticality metrics provide an objective means of assessing safety. However, as existing metrics primarily target longitudinal conflicts, accurately quantifying the risks of lateral conflicts - prevalent in urban settings - remains challenging. This paper proposes the Modified-Emergency Index (MEI), a metric designed to quantify evasive effort in lateral conflicts. Compared to the original Emergency Index (EI), MEI refines the estimation of the time available for evasive maneuvers, enabling more precise risk quantification. We validate MEI on a public lateral conflict dataset based on Argoverse-2, from which we extract over 1,500 high-quality AV conflict cases, including more than 500 critical events. MEI is then compared with the well-established ACT and the widely used PET metrics. Results show that MEI consistently outperforms them in accurately quantifying criticality and capturing risk evolution. Overall, these findings highlight MEI as a promising metric for evaluating urban conflicts and enhancing the safety assessment framework for autonomous driving. The open-source implementation is available at https://github.com/AutoChengh/MEI.

Figures (7)

• Figure 1: Illustration of InDepth. (a) When InDepth is positive, the vehicles will collide unless either (or both) vehicle takes effective evasive action. (b) When InDepth is negative, they won’t collide even without any evasive action.
• Figure 2: Illustration of the safety region. Specifically, when $D_{\text{safe}} = 0$, the safety region is identical to the vehicle body.
• Figure 3: Illustration of ACT and TTC2D calculations. Notably, the closest point pair identified by ACT may not be the actual pair expected to collide, leading it to estimate a collision time nearly half that of TTC2D in this case.
• Figure 4: Distributions of criticality metrics: (top) MEImax, (middle) ACTmin, and (bottom) PET.
• Figure 6: A lateral conflict scenario where the AV turns right and a pedestrian approaches from the right. The minimum ACT is $0.60\,\text{s}$, indicating that ACT may overestimate the actual risk.