Uncertainty-Aware Depth Network for Visual Inertial Odometry of Mobile Robots.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous vehicles and robots.
• Inertial Measurement Units (IMUs) provide cost-effective motion estimation but suffer from noise.
• Visual-Inertial Odometry (VIO) combines cameras and IMUs for robust spatial understanding.

Purpose of the Study:

• To introduce an uncertainty-aware depth network (UD-Net) for improved depth and uncertainty map estimation.
• To develop a novel loss function for training UD-Net.
• To enhance VIO performance by filtering unreliable depth values using uncertainty maps.

Main Methods:

• Developed UD-Net for simultaneous depth and uncertainty map estimation.
• Proposed a novel loss function tailored for UD-Net training.
• Implemented a filtering mechanism using uncertainty maps to refine depth data for VIO.

Main Results:

• UD-Net successfully estimates depth and uncertainty maps.
• The proposed VIO algorithm demonstrates superior performance compared to existing methods.
• Experiments on KITTI and custom datasets validate the effectiveness of the approach.

Conclusions:

• The uncertainty-aware depth network significantly improves VIO accuracy.
• Filtering unreliable depth data based on uncertainty is key to enhancing autonomous system perception.
• The proposed method offers a robust solution for real-world autonomous driving applications.