Onboard Sensors-Based Self-Localization for Autonomous Vehicle With Hierarchical Map

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• Autonomous vehicles require precise localization, but current methods like GPS and SLAM have significant limitations.
• GPS is unreliable in signal-blocked environments, while SLAM methods can be computationally intensive and prone to errors.
• Existing localization techniques struggle with accuracy, robustness, and computational efficiency for real-world autonomous driving.

Purpose of the Study:

• To propose a novel self-localization framework for autonomous vehicles that does not rely on external signals like GPS.
• To achieve centimeter-level localization accuracy with high robustness and instantaneity.
• To develop a lightweight localization approach suitable for embedded computing systems.

Main Methods:

• A homogeneous normal distribution transform algorithm for precise point cloud registration.
• A two-way information interaction mechanism for enhanced localization robustness.
• Hardware and software co-design to create an extremely light-weighted localization system.

Main Results:

• Achieved centimeter-level localization accuracy, meeting autonomous vehicle requirements for instantaneity and robustness.
• Demonstrated a significantly reduced computational footprint compared to traditional LiDAR localization methods.
• Validated effectiveness and advantages through experiments on a public dataset and real-world scenarios.

Conclusions:

• The proposed signal-independent self-localization framework offers a robust and accurate solution for autonomous vehicles.
• The light-weighted design enables operation on embedded systems, overcoming computational limitations of existing methods.
• This approach represents a significant advancement in autonomous vehicle localization technology.