arXiv study: LiDAR dominates fused vehicle localization, radar adds little

Researchers Aleksi Pippuri, Nilusha Jayawickrama, and Risto Ojala released a case study on arXiv (2608.16966) addressing a core challenge in intelligent transportation: accurately localizing vehicles in mixed traffic with both connected and conventional cars. Their proposed multi-observer framework fuses compact object-level detections from a static roadside radar and a dynamic LiDAR-equipped connected vehicle. Using real-world data collected at an urban intersection in Helsinki, Finland, with a separately instrumented target vehicle as ground truth, they benchmarked two extended Kalman filter (EKF) strategies, including an adaptive EKF (AEKF), under nominal conditions, reduced LiDAR update rates, simulated LiDAR occlusions, and different target motion states.

Results show that when LiDAR data is fully available, fusion performance is dominated by LiDAR observations; the less accurate radar detections provide only marginal improvement. The AEKF achieves small gains over a LiDAR-only baseline, and object-level connected vehicle observations remain valuable when shared at reduced update rates. The authors conclude that decision-level fusion gives scenario-dependent benefits rather than guaranteed improvements over a strong single-sensor baseline. Importantly, they released both the dataset and implementation on GitHub, enabling further research in multi-sensor fusion for autonomous and connected vehicles.