A Novel Transformer-Based IMU Self-Calibration Approach through On-Board RGB Camera for UAV Flight Stabilization.

Area of Science:

• Robotics and Control Systems
• Computer Vision
• Sensor Fusion

Background:

• Unmanned aerial vehicles (UAVs) rely on inertial measurement units (IMUs) for accurate pose estimation during flight.
• IMUs, comprising accelerometers and gyroscopes, are susceptible to systematic errors and noise, impacting trajectory precision.
• Traditional hardware calibration methods are often impractical due to equipment requirements or sensor accessibility limitations.

Purpose of the Study:

• To develop a non-invasive, software-based calibration procedure for IMUs in UAVs.
• To mitigate misalignment errors and noise affecting IMU measurements without specialized hardware.
• To improve the overall flight path accuracy and reliability of UAVs.

Main Methods:

• A soft calibration strategy leveraging the UAV's built-in grayscale or RGB camera.
• Implementation of a transformer neural network architecture trained using supervised learning.
• Utilizing pairs of short videos from the UAV camera and corresponding IMU measurements for training.

Main Results:

• The proposed method effectively reduces misalignment errors and noise in IMU data.
• The transformer network accurately learns the relationship between visual data and IMU measurements.
• The soft calibration procedure demonstrated significant potential for enhancing UAV trajectory accuracy.

Conclusions:

• The camera-based soft calibration offers a practical and reproducible solution for IMU error correction in UAVs.
• This approach eliminates the need for specialized calibration equipment, making it widely accessible.
• The method holds promise for advancing autonomous navigation and mission capabilities of UAVs.