Real-Time Sensor Fusion Architecture with Adaptive Weighted Filtering for Autonomous Racing Systems

Engineering notes

Key topics: autonomous driving, object detection, lidar, sensor fusion, carla, radar. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Autonomous racing requires precise location tracking and accurate object detection capabilities to function successfully during both high-speed driving and quick directional changes and simultaneous vehicle operations. The study introduces a sensor fusion system that operates in real time by combining traditional filtering techniques, which include Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Particle Filter (PF), with a new Adaptive Weighted Fusion Filter (AWFF), which changes sensor importance based on vehicle operation and environmental factors. The AWFF system conducts ongoing sensor updates to achieve precise sensor combination results, which operate with a verified average latency of 2.60 milliseconds, well within the 10-millisecond threshold needed for $\mathbf{1 0 0 ~ H z}$ real-time functioning. The fusion system uses LiDAR for creating spatial maps, radar for determining speeds, camera arrays for visual detection, and IMU sensors for measuring dynamic states. The system assessment used synthetic datasets, which were produced through the CARLA simulation environment and F1TENTH simulation tools and Roborace/IAC virtual racing scenarios. The first tests demonstrate a 75.3 % accuracy increase in multi-vehicle positioning results while maintaining 2.60 ms average fusion times, which existing methods struggle to match. The findings demonstrate that AWFF functions as an advanced fusion system that delivers fast results for both future autonomous racing systems and multiple agent systems that operate in unpredictable environments.

Links and sources

• Official / arXiv page

Comments