A research internship at IIT Guwahati spent assembling multi-rotor drones, configuring Pixhawk and Cube Orange flight controllers, and wiring a Raspberry Pi 4 into the loop for real-time vision.
The result: a follower drone that autonomously detects and tracks a manually piloted leader drone using onboard object detection.
A research internship at the Department of Electrical and Electronics Engineering, IIT Guwahati, focused on assembling multi-rotor drone hardware, configuring professional-grade flight controllers, and implementing vision-based autonomy on a Raspberry Pi companion computer — culminating in a drone that autonomously tracks and follows another drone in flight.
A modular build — structural frame, propulsion, power, flight control, and an onboard companion computer for perception — chosen for ease of assembly, debugging, and future scalability.
Structural base for all motors, sensors, and electronics — providing mechanical stability and proper weight distribution.
Brushless motors paired with sized propellers to generate lift and thrust for efficient, stable propulsion.
One ESC per motor, regulating speed from flight-controller signals for precise thrust control.
The Power Distribution Board fed regulated power from a Li-Po battery to ESCs, flight controller, and electronics.
Central processing unit handling sensor fusion, stabilisation, navigation, and control algorithms.
The primary autopilot used for the leader–follower build — interfaced directly with the Raspberry Pi over TELEM2.
Radios for real-time, bidirectional communication with the Ground Control Station via Mission Planner.
Onboard companion computer for real-time video processing, object detection, and autonomous decision-making.
A webcam connected to the Pi to capture the live video feed used for object detection and tracking.
Five stages of hardware integration, each verified before moving to the next — motors first, autonomy last.
Assembly began with mounting the motors and ESCs onto the drone frame. ESCs were connected to the motors and routed to the Power Distribution Board, with care taken over insulation and secure wiring to prevent electrical interference and mechanical damage.
The Pixhawk / Cube Orange was mounted at the frame's centre using vibration-damping mounts to minimise sensor noise, with orientation aligned to manufacturer specifications for accurate attitude estimation.
The battery and power module were integrated to supply regulated power to the flight controller and peripherals. Every power connection was verified with a multimeter before the system was first energised.
The telemetry module was connected to the flight controller to enable bidirectional communication with the Ground Control Station, allowing real-time monitoring of flight parameters through Mission Planner.
The Raspberry Pi 4 was mounted securely with a forward-facing webcam and powered from a regulated power bank. Its GPIO pins were then wired to the Cube Orange's TELEM2 port — TX, RX, and GND — for MAVLink-based serial communication.
Every calibration pass ran through Mission Planner as the Ground Control Station, connected via telemetry or USB — from firmware verification to a validated test flight.
Power on the drone and GCS, connect the flight controller to Mission Planner, and confirm firmware and board configuration.
Place the drone in six orientations and record sensor offsets.
Rotate the drone along all axes, away from magnetic interference.
Calibrate throttle, roll, pitch, and yaw, and assign auxiliary switches.
Synchronise the throttle range for a uniform response across all motors.
Confirm satellite count, fix type, and HDOP before flight.
Configure battery failsafe, RC failsafe, and arming checks.
Assign Stabilise, Loiter, RTL, and Auto-tune, then verify motor direction, propeller orientation, and sensor health.
A successful test flight confirmed the drone was properly calibrated and ready for autonomous experiments.
Troubleshooting hardware, connectivity, and networking issues did more for practical understanding than any tutorial could.
Sparse documentation for the Cube Orange meant small but critical configuration parameters weren't explained anywhere obvious.
Extensive experimentation and cross-referencing official documentation to fill the gaps.
Mission Planner intermittently failed to hold a stable connection, lagging or freezing mid-calibration.
Repeated reconnections and restarts to recover a clean session.
The receiver ignored radio input during calibration.
Traced to improper transmitter–receiver binding — re-binding resolved it.
No remote access to the Pi in the initial setup phase.
SSH and Raspberry Pi Connect, locating the Pi with ping hostname.local — both devices had to share a network.
On USB-C power-up, the Pi sometimes forgot its Wi-Fi credentials and lost internet access, blocking SSH entirely.
Reinstalling the OS repeatedly was too slow. Removing the password and running the target network open resolved the reconnection issue for good.
Two drones fly simultaneously with distinct roles: a manually piloted Leader providing a dynamic target, and an autonomous Follower that processes visual data to maintain relative position — no direct human control of the follower required.
| Cube Orange (TELEM2) | Raspberry Pi GPIO |
|---|---|
| GND | GPIO Pin 9 |
| TX | GPIO Pin 10 |
| RX | GPIO Pin 8 |
With UART communication confirmed, a drone_pursuit.py script tied it all together: a colour-blob tracker locates the leader in each frame, estimates distance from its apparent size, and streams velocity + yaw-rate commands to the Cube Orange over MAVLink — auto-arming, switching to GUIDED_NOGPS, taking off on first detection, and closing the gap while keeping the target centred.
Before wiring detection into flight control, the pipeline was validated independently: a YOLOv8n model reading frames from the webcam, with results pushed out as a low-lag MJPEG stream over the Pi's own HTTP server.
Live inference on the companion computer picked up objects mid-frame with high confidence — the same pipeline later repointed at the leader drone in flight.
"Vision-based autonomous drone following is achievable on low-cost embedded hardware and open-source flight software — a foundation for future work in multi-drone coordination and swarm robotics."
With thanks to Professor Dr. Chayan Bhawal for guidance throughout, to PhD scholar Paraj Ganchaudhuri for constant technical support, and to IIT Guwahati for the opportunity.