DGRO: Doppler Velocity and Gyroscope-Aided Radar Odometry.

Area of Science:

• Robotics and Autonomous Systems
• Sensor Fusion
• Simultaneous Localization and Mapping (SLAM)

Background:

• Autonomous robot navigation requires stable odometry systems.
• 4D millimeter-wave radar offers resilience in adverse conditions and provides position and Doppler velocity data.
• Direct application of LiDAR-based SLAM algorithms is hindered by noise and uncertainty in 4D radar point clouds.

Purpose of the Study:

• To develop a robust SLAM framework fusing 4D radar and gyroscope data for improved robot localization.
• To address the challenges posed by noise and uncertainty in 4D radar point clouds for odometry.
• To enhance the performance of radar-based odometry systems in various environmental conditions.

Main Methods:

• A SLAM framework fusing 4D radar and gyroscope data via graph optimization.
• Utilizing Doppler velocity for ego-velocity estimation and dynamic point removal.
• Implementing a pre-integration factor for ego-velocity and gyroscope data fusion.
• Developing RCS-based point selection and RCS intensity-weighted scan-to-submap registration.

Main Results:

• The proposed DGRO system demonstrates reliable performance and accurate localization.
• Experimental validation on custom and NTU datasets confirms the framework's effectiveness.
• Outperforms traditional 4D radar odometry methods, especially in low-speed scenarios with limited dynamic objects.

Conclusions:

• The DGRO framework successfully fuses 4D radar and gyroscope data for robust robot odometry.
• The system offers enhanced localization accuracy and reliability compared to existing methods.
• The proposed techniques effectively mitigate challenges associated with 4D radar data.