93won/lightweight_vio

What it solves

This project provides a lightweight stereo visual-inertial odometry (VIO) system designed for real-time performance. It addresses the challenge of robust state estimation by implementing a statistical uncertainty learning method to improve how a robot or device tracks its position and orientation in space.

How it works

The system combines data from stereo cameras and inertial measurement units (IMU) using the following technical approach:

• Feature Tracking: Identifies and tracks visual landmarks across frames.
• IMU Pre-integration: Efficiently processes high-frequency IMU data to constrain motion estimates.
• Sliding Window Optimization: Uses the Ceres Solver to optimize the state estimation within a fixed window of recent data to maintain real-time speed.
• Uncertainty Learning: Implements a specific statistical uncertainty learning method to enhance robustness.
• Visualization: Uses Pangolin to provide a real-time visual representation of the estimated trajectory.

Who it’s for

This software is intended for researchers and developers working on robotics, autonomous navigation, and state estimation, particularly those using stereo cameras and IMUs.

Highlights

• Real-time performance: Optimized for low-latency state estimation.
• Ceres Solver integration: Employs professional-grade optimization for visual reprojection and IMU constraints.
• Broad dataset support: Compatible with EuRoC MAV, TUM-VI, and Intel RealSense D435/D435i datasets.
• ROS2 compatibility: Offers a dedicated wrapper for integration into the Robot Operating System 2 ecosystem.