Inspiration

We all wanted to learn how to fly a drone as it's cool to see an autonomous drone to hover.

What it does

The ESP32 drone is in a cubic meter cage, it must hover about 0.5 meters above the ground in the center of the cube and must stay there for 60 seconds.

How we built it

Architecture Camera Feeds → vision.py → position estimate ↓ drone_rc.py ←── controller.py (altitude PID) ↑ config.py (all tunable values) safety.py (e-stop, bounds checking) Project Structure File Purpose main.py Entry point — wires everything together controller.py Altitude PID loop, drone arming/disarming vision.py LED detection, position estimation, Kalman filter safety.py E-stop, signal handlers, attitude bounds checking config.py All tunable constants (gains, thresholds, targets) drone_rc.py Provided comms library — do not modify Setup pip install -r requirements.txt Connect to the drone's Wi-Fi network (192.168.4.1) before running.

Usage Test 1 — Comms + basic hover (no CV):

python main.py Test 2 — Full CV hover:

python main.py --cv

optionally specify camera indices if defaults (0, 1) are wrong:

python main.py --cv --cam0 0 --cam1 2 LED calibration (run before Test 2):

python -c "from vision import led_calibration; led_calibration()" Adjust the HSV trackbars until only the drone LED is visible in the mask window, then update LED_HSV_LOWER / LED_HSV_UPPER in vision.py.

E-Stop Spacebar — primary e-stop Ctrl+C — also routed through e-stop (won't leave motors running) E-stop sends mode0 to the drone and exits the control loop cleanly.

Tuning All tunables are in config.py. Key values to adjust during testing:

Parameter What to do BASELINE_THRUST Increase from 100 until drone lifts, note hover value PITCH_GAIN_P/I/D Offset drone's internal PID if oscillating or drifting ALT_P Increase if altitude response is sluggish ALT_D Increase to dampen altitude oscillation / fight turbulence CONTROL_HZ Keep at 20 — higher values stress drone's gyro update budget Test 1 Checklist Connected to drone Wi-Fi python main.py runs without hanging on comms check Spacebar triggers e-stop before any motors spin Mode 2 sets correctly (get_mode() returns expected value) BASELINE_THRUST tuned to approximate hover point

Challenges we ran into

We ran into multiple challenges as we struggled to get the computer vision done. The main challenge was getting the communication between the IMU of the drone and computer vision.

Accomplishments that we're proud of

We are proud to at least make the drone fly straight in the air and also managed to get the computer vision partially working.

What we learned

We learned that it takes lots of attempts to get a drone to fly the way we wanted. There are lots of steps to do to get a drone to hover 0.5 meters above the ground.

What's next for skyforge

The next step is continuing to learn how drones fly.

Built With

• python