Probe-Drogue Relative Vectoring for Autonomous Aerial Refueling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sensing technologies for autonomous aerial refueling, such as GPS and inertial navigation systems, are unreliable and fail to provide accurate, real-time pose estimation for docking between aircraft, while existing vision algorithms do not meet the precision and speed requirements for safe and efficient air-to-air refueling.
Innovation Solution
A computer vision system using dual object detection and relative vectoring methods, including You Only Look Once (YOLO) for object detection and Solve PnP for pose estimation, enables precise localization of the probe and drogue without relying on extrinsic camera properties, achieving less than 3 cm error in position and less than 1 degree in orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If GPS and inertial navigation systems are used for autonomous aerial refueling, then the system can operate autonomously, but the measurement precision and reliability of pose estimation deteriorates
Solution Approach 1:
The patent replaces GPS and inertial navigation systems with a computer vision-based optical measurement system. The system uses cameras to capture images of the drogue and probe, then applies computer vision algorithms to extract pose information, substituting mechanical/navigation systems with an optical-field-based solution that achieves superior measurement precision.
Solution Approach 2:
The patent introduces computer vision algorithms as an intermediary between the camera images and the pose estimation. The vision algorithms process the images to extract geometric features and compute the relative pose, serving as a mediator that translates optical information into precise spatial relationships without relying on GPS or inertial sensors.
2Productivity
If existing vision algorithms are used for dual object localization, then the system can detect objects, but the measurement precision and processing speed deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-defining the geometric relationships and coordinate transformations between the camera, drogue, and probe. The system establishes predetermined measurement models and transformation matrices that enable direct computation of pose from image coordinates, avoiding iterative optimization and achieving both high speed and high precision.
3Reliability
If extrinsic camera properties are used for pose estimation, then the system can compute relative position, but the reliability deteriorates due to calibration errors and occlusions
Solution Approach 1:
The patent extracts and eliminates the dependency on extrinsic camera properties from the pose estimation process. By formulating the measurement model to be independent of camera calibration parameters, the system removes the source of calibration errors and improves reliability without requiring complex calibration procedures.
Data Source
AI summary
A method of determining an object-to-object vector. The method includes providing a camera on one of a first object having a first object docking member and a second object having a second object docking member. The camera captures a 2D image including the first object docking member and the second object docking member. A plurality of 2D image points are identified on the 2D image and matched to some of 3D features of the first object docking member and some of the 3D features of the second object docking member. Camera frame first and second object vectors are subtracted to determine a camera frame first object to second object vector.