Computer Vision-Guided Autonomous Drone Navigation & Flight Simulation
2025YOLO family · MATLAB sim · ML pipeline · path planning
Vision models for gate detection and pose estimation coupled with MATLAB flight simulation and closed-loop control for autonomous racing-style quadcopter navigation.
Overview
This system combines advanced computer vision models with robust flight dynamics simulations to enable autonomous high-speed quadcopter navigation. By training deep learning frameworks to detect racing gates and estimating real-time poses, the system feeds visual data into a closed-loop control pipeline developed in MATLAB to optimize path-planning and flight stability in dynamic environments.
Key features
- Multi-architecture detection: YOLOv5, YOLOv7, and YOLO Pose tuned for high frame-rate gate tracking and 3D localization.
- MATLAB simulation framework integrating 6-DoF dynamics with tuned PID loops for trajectory validation.
- Python automation for XML extraction, preprocessing, and feature extraction to speed dataset prep and training quality.
- Guidance / path-planning translating detections into smooth, collision-free trajectories inside the simulator.
Control strategy
The ML pipeline separates heavy preprocessing from training to keep GPUs saturated. In simulation, fast inner-loop PID control runs decoupled from slower vision updates to preserve stability. Feedback between geometry modules and the dynamics solver is structured to limit latency accumulation and avoid divergence during aggressive maneuvers.
Gallery
