Implementation and analysis of a parallel kalman filter algorithm for lidar localization based on CUDA technology

Area of Science:

• Robotics and Autonomous Systems
• Sensor Fusion
• Computational Geometry

Background:

• Navigation satellite systems (e.g., GPS) are prone to failure due to environmental and technical issues, impacting autonomous driving localization accuracy.
• Lidar (Light Detection and Ranging) offers an alternative localization technology, but its integration with existing systems requires optimization.
• Kalman filters enhance Lidar measurement accuracy by accounting for noise and inaccuracies.

Purpose of the Study:

• To develop a computationally efficient Lidar localization algorithm for autonomous driving.
• To improve the speed of Lidar-based localization without compromising accuracy.
• To address the limitations of satellite navigation in challenging conditions.

Main Methods:

• Proposed a parallel Kalman algorithm implemented in three-dimensional space.
• Focused on parallelizing the Kalman localization algorithm itself, not map generation.
• Utilized CUDA for accelerating the Kalman filter with Lidar data.

Main Results:

• Achieved a 3.8x speedup in Lidar localization computation.
• Maintained localization accuracy at 3% in both parallel and non-parallel implementations.
• Demonstrated the effectiveness of the parallel Kalman algorithm for real-time decision-making.

Conclusions:

• The parallel Kalman algorithm significantly enhances the computational speed of Lidar localization.
• This approach offers a practical solution for real-time autonomous driving, especially with large Lidar datasets.
• The method provides a reliable and accurate localization solution, overcoming satellite navigation vulnerabilities.