Roadside LiDAR Assisted Cooperative Localization for Connected Autonomous Vehicles

TL;DR

The paper tackles the challenge of precise vehicle localization when onboard self-localization and HD map features are insufficient. It proposes a roadside LiDAR assisted cooperative localization approach leveraging V2I communication to refine the target vehicle’s position, complemented by a bounding box size correction that uses known vehicle dimensions to improve center localization, achieving centimeter-level accuracy in realistic ranges. Through a feasibility study in a Unity-based simulator and a realistic Autoware-AWSIM co-simulation in a Tokyo district, the method outperforms traditional NDT scan matching under several conditions, with notable gains when using higher-end LiDAR models. The work demonstrates a practical path toward reducing reliance on dense HD maps and improving urban autonomous driving safety, while outlining future work on multi-vehicle scenarios, occlusion handling, and LiDAR placement optimization.

Abstract

Advancements in LiDAR technology have led to more cost-effective production while simultaneously improving precision and resolution. As a result, LiDAR has become integral to vehicle localization, achieving centimeter-level accuracy through techniques like Normal Distributions Transform (NDT) and other advanced 3D registration algorithms. Nonetheless, these approaches are reliant on high-definition 3D point cloud maps, the creation of which involves significant expenditure. When such maps are unavailable or lack sufficient features for 3D registration algorithms, localization accuracy diminishes, posing a risk to road safety. To address this, we proposed to use LiDAR-equipped roadside unit and Vehicle-to-Infrastructure (V2I) communication to accurately estimate the connected autonomous vehicle's position and help the vehicle when its self-localization is not accurate enough. Our simulation results indicate that this method outperforms traditional NDT scan matching-based approaches in terms of localization accuracy.

Figures (9)

• Figure 1: Setup of the Pilot Experiment to Generate Heat Map
• Figure 2: Error and number of points when using the original L-shape fitting algorithm on VLP-16
• Figure 3: Error of the Center Point in Distance (m) after bounding box size correction.
• Figure 4: The bird view of the Nishishinjuku Area in AWSIM. The pink arrow is where we conducted the realistic experiment, and the blue dot is where we put the roadside LiDAR.
• Figure 5: Background filtering. The blue points are background points; the white points are vehicle points.