Core Concepts
Integrating Vehicle-to-Vehicle (V2V) communication with traditional sensor fusion can enhance the reliability and resilience of autonomous vehicle perception systems, particularly in complex driving scenarios with occlusions.
Abstract
The proposed system integrates detections from local sensors (camera and radar) with Vehicle-to-Vehicle (V2V) Basic Safety Messages (BSMs) to create a more comprehensive and robust track management system. The key innovations include:
Creation of independent priority track lists that fuse detections from local sensors and validate them through V2V communication. This allows for more flexible and resilient thresholds for track management, especially in scenarios with occlusions.
Consideration of the potential for falsification of V2X signals, which is addressed through an initial vehicle identification process using detections from perception sensors before incorporating the V2V data.
Simulation of complex driving scenarios, including a 4-way intersection with unprotected left turns, to evaluate the performance of the V2V-enabled sensor fusion system against traditional local sensor fusion.
The experimental results demonstrate the improved accuracy and robustness of the proposed system compared to local sensor fusion alone, as measured by key tracking metrics such as GOSPA, Missed Target Error, False Track Error, and Switching Error. The integration of V2V data helps mitigate the limitations of perception sensors, particularly in occluded environments, and enhances the reliability and efficiency of autonomous vehicle systems.
Stats
The average GOSPA metric values over the simulation interval were:
Local sensor fusion tracks: 56.12
V2V tracks: 7.517
Priority V2V-enabled sensor fusion tracks: 48.62
The average Missed Target Errors were:
Local sensor fusion tracks: 30.0
V2V tracks: 0.0
Priority tracks: 21.2
The average False Track Errors were:
Local sensor fusion tracks: 47.43
V2V tracks: 0.0
Priority tracks: 42.43
The average Switching Errors for all three tracking systems was 0.0.
Quotes
"Utilizing V2V information not only extends the operational range of the track management systems but also improves the resilience of safety-critical features such as Autonomous Intersection Navigation (AIN)."
"The capability to significantly lower error rates in these metrics is crucial to marketing ADAS features in vehicles as both highly effective and safe for users."