MAV Localization in Large-Scale Environments: A Decoupled Optimization/Filtering Approach.

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Micro aerial vehicle (MAV) localization in large-scale environments is challenging.
• Existing sensor fusion algorithms often suffer from high latency and computational complexity.
• Accurate and low-latency localization is critical for MAV operations.

Purpose of the Study:

• To develop a sensor fusion algorithm for GPS-aided MAV localization that guarantees high accuracy and minimal system delay.
• To propose a linear optimal state estimation approach and an immediate metric-scale recovery paradigm.
• To enable robust MAV pose estimation using vision sensors and low-rate GPS data.

Main Methods:

• Linear optimal state estimation to avoid complex, high-latency calculations.
• Metric-scale recovery using noisy, low-rate GPS measurements.
• Optimization/filtering-based methodology treating the camera as a 'black-box' pose estimator.
• Consideration of sensor measurement uncertainty constraints for GPS-limited scenarios.

Main Results:

• The proposed strategy enables vision sensors to quickly bootstrap arbitrarily scaled poses.
• The method effectively recovers from drifts common in vision-based algorithms.
• The approach maintains low computational complexity, suitable for long-term MAV operations.
• Demonstrated superior performance in trajectory estimation accuracy and system latency compared to state-of-the-art algorithms.

Conclusions:

• The developed sensor fusion technique significantly enhances MAV localization accuracy and reduces latency.
• The approach provides a computationally efficient and robust solution for MAVs operating in large-scale landscapes.
• This method offers a reliable method for MAVs to leverage vision and GPS data effectively, even with noisy or limited GPS signals.