An Object-Centric Hierarchical Pose Estimation Method Using Semantic High-Definition Maps for General Autonomous...

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Autonomous Systems

Background:

• Robust autonomous driving systems require precise vehicle pose estimation.
• Current methods using Real-Time Kinematic (RTK) sensors struggle in GPS-denied environments like indoors or areas with signal interference.
• Inaccurate pose estimation hinders the development of reliable autonomous vehicles.

Purpose of the Study:

• To develop a more universal and robust method for vehicle pose estimation.
• To overcome the limitations of RTK sensors in challenging environments.
• To enhance the stability and reliability of autonomous driving systems.

Main Methods:

• Leveraging semantic high-definition (HD) maps with registered objects.
• Creating object-centric features from the HD map.
• Recognizing vehicle location using these object-centric features.
• Estimating vehicle pose based on the recognized location.

Main Results:

• The proposed method significantly improves vehicle pose estimation precision in environments with poor RTK signal reception.
• Enhanced robustness and stability of autonomous driving systems in challenging scenarios.
• Demonstrated effectiveness through both simulation and real-world experiments.

Conclusions:

• The object-centric approach using HD maps provides a viable solution for accurate vehicle pose estimation where RTK fails.
• This method contributes to more reliable and widespread adoption of autonomous driving technology.
• Further research can explore integration with other sensor modalities for even greater robustness.