Continuous Error Map-Aided Adaptive Multisensor Integration for Connected Autonomous Vehicles in Urban Scenarios

Abstract

Precise multi-sensor integrated positioning is essential for autonomous vehicles (AVs) in urban environments. One of the key challenges in multi-sensor fusion is accurately estimating the weights of heterogeneous sensor data. With the emergence of Cellular Vehicle-to-Everything (C-V2X) technology and smart roadside infrastructure (RSIs), these systems can collaborate to provide enhanced and reliable services to connected vehicles. Motivated by this, our research explores the use of sensor error maps for heterogeneous sensor measurements under varying environmental conditions to improve the positioning accuracy of connected AVs in complex urban areas. We propose a multi-sensor integrated positioning system that utilizes error map information generated by sensor-rich connected autonomous vehicles (CAVs). This error information is shared with RSIs and then distributed to nearby CAVs within the same region. Sensors with higher estimated errors are assigned lower weights, as determined by the error maps. To validate the proposed approach, we conducted experiments both day and night in a realistic simulation environment as well as in the Hong Kong C-V2X testbed. The results demonstrate that the use of continuous error maps significantly enhances the performance of multi-sensor integrated positioning. The data and implementation of our system are available at https://github.com/DarrenWong/continuous error map.

Problem Addressed

Accurately estimating the weights of heterogeneous sensor data is a key challenge in multi-sensor fusion, because sensor errors vary with environmental conditions across complex urban areas.

Key Contributions

• Continuous error map estimation from onboard sensors with varying confidence across urban environments
• Error information sharing between connected autonomous vehicles and roadside infrastructure via C-V2X
• Adaptive multi-sensor integration assigning lower weights to higher-error sensors based on error maps
• Day and night validation in simulation and the Hong Kong C-V2X testbed, with open-sourced implementation

Conclusions

Continuous error maps, shared in real time via roadside infrastructure, enable dynamic sensor weighting that significantly enhances multi-sensor integrated positioning. The approach is validated in high-fidelity simulation and the real-world Hong Kong C-V2X testbed under complex urban conditions, both day and night.