Enabling robust relative localization and control between multiple UAVs by fusing RTK-GPS, UWB ranging, and vision-based pose estimation through an EKF framework
My ISEF project: Solving the challenge of autonomous aerial docking through multi-sensor fusion
I have developed a multi-sensor fusion system for robust relative localization and autonomous mid-air docking between two UAVs. My system fuses RTK-GPS, UWB ranging, and vision-based pose estimation through an EKF framework, enabling precise aerial docking, cooperative UAV formations, and operations in GPS-degraded environments. This is my original research project for the International Science and Engineering Fair (ISEF).
Acts as the docking platform with 4 UWB anchors and AprilTag markers
Performs autonomous approach and docking with sensors and electromagnet
Each sensor operates at different ranges and precision levels, enabling a seamless docking sequence
Global Positioning System provides universal positioning with no distance limitations. Allows initial approach to the mother UAV's general vicinity.
Ultra-Wideband ranging measures direct distances to 4 UWB anchors on the mother UAV, providing 3D relative position estimation.
AprilTag-based visual pose estimation provides the highest precision alignment for final docking phase with millimeter-level accuracy.
Extended Kalman Filter fuses all sensor data into a continuous, stable state estimate, ensuring smooth transitions between sensor handover phases.
Multi-stage control logic I developed ensures safe and precise aerial docking
My child UAV uses RTK-GPS to navigate to the general vicinity of the mother UAV (within ~10m). GPS provides coarse positioning without distance limitations.
Within 10m range, my UWB system activates. The child UAV measures distances to 4 anchors on the mother UAV and computes relative 3D position with ~10cm accuracy, gradually aligning and approaching.
Within 0.5m, my vision system identifies AprilTag markers on the mother UAV. Visual algorithms decode precise relative pose with ~1cm accuracy for final alignment.
Once aligned, the child UAV's electromagnet activates and locks onto the steel docking plate of the mother UAV, completing the autonomous mid-air docking without human intervention.
Overcoming real-world obstacles to achieve autonomous mid-air docking
Wind gusts and rotor downwash from both UAVs create unpredictable aerodynamic disturbances during approach.
Sensor noise and integration errors cause relative position estimation to drift over time, making precise docking impossible.
No single sensor provides full coverage: GPS is imprecise, UWB has limited range, vision fails at distance.
Processing multiple sensor streams and running EKF at 100Hz requires significant computational resources.
Electromagnet requires sub-centimeter alignment for successful connection. Misalignment causes docking failure.
MAVLink messages between companion computer and flight controller introduce control loop delays.
ISEF-aligned risk assessment and procedures for safe testing
My project involves typical UAV risks that must be identified and controlled.
Components and activities that require extra caution during development.
Steps I take to reduce risk during testing and operation.
Safe handling and disposal of any damaged components.
Guidelines I follow to ensure safe and compliant testing.
Bridging the gap between theory and real-world autonomous aerial operations
Key differentiators from existing approaches
Unlike single-sensor systems, I fuse GPS, UWB, and vision seamlessly, each operating in its optimal range. This is the only practical solution for full-scale docking.
Real UAVs with PX4 autopilot, not simulation. Demonstrates actual aerodynamic challenges and validates algorithms in real-world conditions.
Built on PX4 (open-source) and ESP32 (affordable). Anyone can replicate, modify, and extend the system without expensive proprietary hardware.
Mechanical locking via electromagnet provides reliable connection. Alternative methods (mechanical claws, visual servoing) are more complex and less robust.
Explicitly designed for aerodynamic disturbances. High-rate control loop (100Hz) and predictive compensation outperform standard approaches.
~1m accuracy. Insufficient for docking. My system achieves <1cm final accuracy.
Fail at >2m range. I use GPS/UWB for approach, vision only for docking.
~10cm accuracy not enough for magnetic docking. Vision provides sub-centimeter precision.
Misses real aerodynamic effects. My system handles actual wind and downwash.
Expensive and closed. I use affordable hardware with open-source software.
Requires skilled pilot. My system is fully autonomous from takeoff to lock.
From industrial operations to advanced research, this technology addresses critical needs
Most direct and high-impact application
Open-source flight control meets custom sensor fusion
Reliable autopilot hardware ensuring stable flight
DW1000-based Ultra-Wideband ranging (4 anchors + 1 tag)
Machine vision with AprilTag detection
High-performance computing for sensor fusion and control
Secure mechanical connection for mid-air docking
Open-source flight control software - handles basic flight operations
Multi-sensor state estimation for smooth, continuous positioning
Linear least squares solving for position from UWB distances
Fiducial marker system for high-precision visual pose estimation
Communication between companion computer and flight controller
While single-sensor approaches fail in different scenarios, my multi-sensor fusion system provides robust performance across all stages of docking. GPS ensures I can find the mother UAV from anywhere. UWB provides stable convergence when GPS is insufficient. Vision guarantees millimeter-to-centimeter alignment for final docking. The EKF framework keeps the process continuous and controllable. This architecture directly targets ISEF-level research needs for robust, real-world UAV coordination.