DeepFlyer¶
Learn Drone AI Through Reinforcement Learning
DeepFlyer is a browser-based platform for learning drone control, autonomous navigation, and AI training, all inside a safe cloud simulation powered by Gazebo, PX4, and ROS 2. No downloads or hardware needed.
What You Can Do¶
🚁 Fly a Drone Keyboard controls, live video
🗺 Plan Missions Waypoint autopilot
🧩 Write Logic If/else obstacle avoidance
📐 Tune a Controller PID gains, 3D trajectory
🧠 Train an AI Agent Live reward graph, leaderboard
Five Activities, One Learning Path¶
Each activity builds on the last, taking you from basic flight to running your own AI experiment.
1 Manual Control Beginner - 15 min
2 Waypoint Nav Beginner - 15 min
3 Obstacle Avoidance Beginner - 20 min
4 PID Tuning Intermediate - 30 min
5 RL Training In Progress
Where Do I Start?¶
| If you want to... | Go here |
|---|---|
| Create an account and fly for the first time | Quick Start |
| Learn the site navigation and session system | The Interface |
| Browse all five activities | All Activities |
| Check your stats, badges, and ranking | Dashboard |
| Read about the tech stack | Technical Docs |
Each activity start takes about 30 seconds
Every time you launch an activity, the platform spins up a fresh simulation container for you. This takes around 30 seconds. Once the status dot turns green, wait an additional 1 to 2 minutes before arming the drone to give the simulation environment time to fully load.