CoDrive-MLF: Empirical Validation of Cooperative Driving via a Multi-Layer V2V Evaluation Framework

초록

Standalone Automated Driving Systems (ADS) face inherent limitations in safety and efficiency due to sensor occlusion and uncertainty in predicting other vehicles’ intentions. Vehicle-to-Vehicle (V2V) cooperation has been proposed to address these challenges, yet empirical studies directly comparing cooperative and standalone ADS under identical real-road conditions remain limited. To address this gap, this paper presents CoDrive-MLF, an integrated cooperative driving and evaluation framework that pairs With Cooperation (WC) and Without Cooperation (WOC) runs using identical vehicles, controllers, and planners. Cooperation benefits are quantified through four layers: uncertainty, Quality-of-Service (QoS)–safety coupling, behavior patterns, and performance. Field experiments in Cooperative Lane Merge (CLM) and Emergency Trajectory Alignment (ETrA) scenarios were conducted on public roads in Incheon, Korea. Results show that V2V cooperation reduced position RMSE by 95.4% and state uncertainty by 0.67–2.10 nats compared to LiDAR-only tracking; when uncertainty reduction exceeded 0.8 nats, Time-to-Collision (TTC) below 2 seconds dropped to 0% and Deceleration Rate to Avoid a Crash (DRAC) decreased by up to 99.58%. These findings provide real-road quantitative evidence that V2V cooperation enhances safety through reduced uncertainty and earlier situational awareness. This work addresses a critical gap in cooperative driving validation. © 2020 IEEE.

키워드