Detailed Introduction
NeuralFlight is an open-source framework for drone control that combines computer vision (Mediapipe-based) with motor-imagery EEG classification. It enables control of simulated drones using hand gestures, head movements, or imagined actions, all without expensive hardware. The project uses PyTorch for model training and provides a simulator, runnable demos, and example notebooks for rapid prototyping and research.
Main Features
- Multi-modal control: fist-following hand gestures, head-pose control, and EEG-based motor imagery control.
- Modern ML stack: PyTorch-based models (EEGNet with residual connections), real-time inference, and pretrained checkpoints.
- Simulation-first demos: physics-based simulator and visualization let users develop and test without physical drones.
Use Cases
- BCI research and rapid prototyping of EEG-based control algorithms.
- Accessibility: alternative control methods for users with motor impairments.
- Education: teaching signal processing, deep learning, and robotics using hands-on demos.
Technical Features
- EEG pipeline with dataset integration (PhysioNet Motor Movement/Imagery) and bandpass filtering.
- Compact neural architectures (~10K parameters) with residual connections for efficient training.
- Mediapipe-based hand and face tracking, temporal smoothing, and configurable gesture thresholds for stability.