LiDAR-Based GNSS Denied Localization for Autonomous Racing Cars

Area of Science:

• Robotics
• Autonomous Systems
• Vehicle Dynamics

Background:

• Autonomous vehicles offer significant advancements in transportation efficiency and safety.
• Accurate localization is critical for autonomous vehicles, especially race cars operating at the limits of friction.
• Global Navigation Satellite Systems (GNSS) can be unreliable in certain environments.

Purpose of the Study:

• To develop and evaluate a novel, GNSS-independent localization architecture for autonomous race cars.
• To address the critical need for precise pose estimation in high-speed, dynamic driving scenarios.
• To improve the robustness and accuracy of localization systems in autonomous racing.

Main Methods:

• Implemented a localization architecture combining two multi-rate Extended Kalman Filters.
• Extended a state-of-the-art laser-based Monte Carlo localization (MCL) approach with environmental and contextual knowledge.
• Compared the proposed method against a standard state-of-the-art implementation.

Main Results:

• The proposed architecture demonstrated robustness against the 'kidnapping problem' common in particle filter localization.
• Achieved smooth, high-rate pose estimation essential for real-time control.
• Pose error varied with velocity, averaging 0.1m laterally at 60 km/h and 1.48m at 200 km/h; longitudinally, errors ranged from 1.9m at 100 km/h to 4.92m at 200 km/h.

Conclusions:

• The developed GNSS-independent localization system is effective for autonomous racing applications.
• The architecture provides reliable and accurate pose estimation, outperforming existing methods in specific scenarios.
• This research contributes to the advancement of autonomous vehicle control and safety in extreme conditions.