FlockRL
Autonomous drone path planning via sim-to-real transfer of reinforcement learning policies.
Project overview
Problem
Optimizing and automating flight paths in relatively static environments like factory floors.
Solution
Training optimal navigation policies in simulation for deployment on resource-constrained drones.
Objective
Investigate sim-to-real transfer of learned drone trajectories under open-loop deployment.
Future objectives
Expand to dynamic obstacle environments and real-time path adjustment. Scale to multi-drone swarm navigation.
Tech stack
RL Training
- Gymnasium — environments
- Stable-Baselines3 — RL algorithms
- TensorBoard — training monitoring
- Plotly — visualization
Hardware
- ESP32-based WiFi flight controller
Demo
Trained policy navigating a structured environment in simulation.
LinkedIn
