Skip to content

HARDWARE.md

Latest commit

 

History

History
90 lines (71 loc) · 5.1 KB

HARDWARE.md

File metadata and controls

90 lines (71 loc) · 5.1 KB
Raw

Hardware

FlyDrones talks to every drone through one small interface (drones/base.py): takeoff(), send(FlightCommand), telemetry(), frame(), land(). A FlightCommand has throttle (vertical speed), yaw (yaw rate), forward, lateral, each in −1..1. The drone's own flight controller turns that into motor speeds and keeps the airframe level.

Status: the simulator path is tested end to end. The Tello, Crazyflie, MAVLink and ESP32 adapters follow the official SDKs but have not been flight-tested by the authors yet. Please open an issue with logs (--log flight.csv) when you try one.

DJI / Ryze Tello (recommended first drone)

  • Why: built-in altitude hold and optical-flow positioning, Wi-Fi video, prop guards, 80 g.
  • Install: pip install -e ".[tello,gestures]"
  • Connect: join the TELLO-XXXXXX Wi-Fi network from your laptop.
  • Run: flydrones fly --drone tello --config configs/tello.yaml --input both --live --send
  • Mapping: send_rc_control(lateral, forward, throttle, yaw) scaled to ±60 % stick by default.
  • Eyes: the Tello video feed goes into the retina (optic flow and looming). Your hand in front of the laptop webcam adds illusions. --input camera uses only the drone camera, --input gesture only the hand.
  • Kill switch: Ctrl+C lands. TelloDrone.emergency_stop() cuts motors (the drone falls).
  • Tip: Tello needs light and a textured floor to hold position. Video over Wi-Fi lags 100-200 ms, which slows the looming reflex.

Bitcraze Crazyflie 2.1 (+ Flow deck v2)

  • Why: 27 g, open firmware, safest indoor platform.
  • Install: pip install -e ".[crazyflie,gestures]" and a Crazyradio PA/2.0 dongle.
  • Run: flydrones fly --drone crazyflie --uri radio://0/80/2M/E7E7E7E7E7 --config configs/crazyflie.yaml --input gesture --send
  • Mapping: hover setpoints (vx, vy, yaw_rate, z). The brain's throttle is integrated into a height target, so the Crazyflie's estimator holds altitude between decisions.
  • Eyes: no video camera. Use the webcam hand (or add an AI deck and write a frame() method).
  • Arming: newer firmware needs an arming request, the adapter sends it.

ArduPilot / PX4 (MAVLink)

  • Start in SITL. ArduPilot: sim_vehicle.py -v ArduCopter --console --map. PX4: make px4_sitl gz_x500.
  • Install: pip install -e ".[mavlink]"
  • Run: flydrones fly --drone mavlink --mavlink udpin:0.0.0.0:14550 --autopilot ardupilot --config configs/mavlink_sitl.yaml --input gesture --send
  • Serial telemetry radio: --mavlink COM5,57600 (Windows) or /dev/ttyUSB0,57600.
  • Mapping: SET_POSITION_TARGET_LOCAL_NED in body frame, velocity + yaw rate (type_mask = 1479). ArduPilot uses GUIDED, PX4 uses OFFBOARD (a setpoint stream is sent before switching).
  • Telemetry used: LOCAL_POSITION_NED (altitude, geofence), ATTITUDE (yaw rate → halteres), SYS_STATUS (battery).
  • Outdoors only, with a real RC transmitter able to switch to LOITER/LAND at any time.

Betaflight / INAV quad via ESP32 bridge

For FPV-style quads without a companion computer.

laptop (fly brain) ──Wi-Fi UDP──► ESP32 ──UART MSP──► flight controller ──► ESCs
  • Parts: any ESP32 dev board (or M5Stack Atom), 3 wires, a Betaflight/INAV FC with a free UART.
  • Firmware: open firmware/esp32_msp_bridge/esp32_msp_bridge.ino in Arduino IDE (ESP32 core ≥ 2.0), change AP_PASS, flash.
  • Wiring: ESP32 GPIO17 → FC RX, GPIO16 ← FC TX, GND ↔ GND.
  • Betaflight Configurator: Ports → MSP on that UART. Receiver → "MSP RX input". Modes → ARM on AUX1, ANGLE always on. Failsafe → stage 2 "Land".
  • Laptop: join FlyDrones-Bridge Wi-Fi, then flydrones fly --drone esp32 --config configs/esp32_betaflight.yaml --input gesture --send.
  • Protocol: FD1,seq,arm,thr,yaw,pitch,roll at up to 50 Hz, values −1000..1000. The ESP32 stops sending RC frames after 300 ms without a packet, so the FC's own RX-loss failsafe takes over.
  • Warning: Betaflight has no altitude hold. Throttle here is a stick offset around HOVER_PWM, which you must calibrate. This is the hardest and most dangerous path. Props off until everything is verified, then fly in a cage or over a net.

Your own drone

Subclass Drone:

from flydrones.drones.base import Drone
from flydrones.safety import Telemetry

class MyDrone(Drone):
    name = "mine"
    has_camera = False
    def takeoff(self): ...
    def land(self): ...
    def send(self, cmd): ...          # cmd.throttle, cmd.yaw, cmd.forward, cmd.lateral in -1..1
    def telemetry(self): return Telemetry(alt_m=..., yaw_rate_dps=..., battery_pct=..., flying=True)

Then run it with flydrones.runtime.run_realtime(Pilot(brain, MyDrone(), cfg)).

Laptop

  • Any 4-core CPU from the last few years runs MiniFly and sensorimotor cores in real time.
  • The full MaleCNS brain needs 8 GB RAM. Speed depends on the CPU; check with flydrones bench --brain ....
  • A webcam for gestures; good, even lighting helps both MediaPipe and the OpenCV fallback.