Challenges
What went wrong and what we did
Eight problems we solved along the way, and five limits we are still working on. The problems taught us more than the wins.
01Starting with no drone experience
It was our first aerospace project, and all five of us were new to it.
Starting with no drone experience
It was our first aerospace project, and all five of us were new to it.Symptom
Wrong motor, ESC or propeller choices could have doomed the build before any code was written.
Cause
The gap was knowledge of thrust, current draw and which parts work together.
How we found it
We asked seniors, mentors and robotics teams at our university and others about real failures.
Fix
Their advice shaped a parts list where the frame, motors, ESCs and battery suit each other.
Lesson
Ask people who have failed already. It is the cheapest test there is.
Go deeper
02Parts we could not buy locally
Suitable wire and connectors were not sold in the local market.
Parts we could not buy locally
Suitable wire and connectors were not sold in the local market.Symptom
High-current silicone wire and vibration-proof connectors were unavailable, and importing cost too much and took too long.
Cause
Regional supply limits against a fixed academic deadline.
How we found it
We tested locally available connectors for current capacity, grip under vibration and ease of soldering.
Fix
We kept the safest connectors and harvested quality wire from cheap assemblies for the power and sensor harnesses.
Lesson
Check real capability rather than the label, and look at what is already inside cheap products.
Go deeper
03An ESP32 cannot be a flight controller
Our first brain could not keep steady timing.
An ESP32 cannot be a flight controller
Our first brain could not keep steady timing.Symptom
Control timing was unstable during early firmware prototypes.
Cause
Wi-Fi and Bluetooth stacks and the FreeRTOS scheduler added unpredictable jitter and latency.
How we found it
We reviewed how established systems such as Pixhawk and Betaflight are built, and found they rely on STM32 chips.
Fix
We redesigned around the STM32H743, with its hardware floating point, and repurposed the ESP32 for telemetry.
Lesson
Pick a chip for its timing guarantees as well as its clock speed.
Go deeper
04I2C bus freezing as motors spin up
Sensor data stopped whenever the ESCs drew hard.
I2C bus freezing as motors spin up
Sensor data stopped whenever the ESCs drew hard.Symptom
The I2C bus to the sensors randomly froze during motor spin-ups, which would crash a drone in flight.
Cause
An oscilloscope showed voltage dips and high-frequency noise on shared power rails during high-current draw.
How we found it
We probed the rails and bus lines with an oscilloscope while spinning the motors.
Fix
We soldered ceramic and electrolytic capacitors across VCC and GND at the sensors and STM32 rail, and later added a software bus recovery.
Lesson
Fix noise at its source first. Software recovery is a backup.
Go deeper
05Clone chips under the same name
The same part name could hide different silicon.
Clone chips under the same name
The same part name could hide different silicon.Symptom
Sensor boards sold under one name contained different chips, with different addresses and registers.
Cause
Regional vendors and batch variation, for example HMC5883L, QMC5883L and Si24R1 clones.
How we found it
Fixed drivers crashed or read garbage on a chip that did not match the datasheet.
Fix
At boot, the firmware scans the bus, identifies the chip and loads the correct driver and scaling.
Lesson
Assume parts will vary, and detect them at boot.
Go deeper
06No budget for a test bench
Free-flight tuning risked destroying the drone.
No budget for a test bench
Free-flight tuning risked destroying the drone.Symptom
Starting PID values are only estimates, so a free-flight mistake could end the project.
Cause
Professional multi-axis test benches were far beyond our budget.
How we found it
We looked at what such benches actually do: restrict movement while allowing rotation.
Fix
We built wooden rigs that lock the drone in place and let it rotate on one axis, so gains could be pushed safely to their limits.
Lesson
You often need the principle behind expensive equipment, rather than the equipment itself.
Go deeper
07Uneven thrust from budget motors
The drone drifted, and it was not the software.
Uneven thrust from budget motors
The drone drifted, and it was not the software.Symptom
After static tuning, the drone drifted in pitch and roll on every liftoff attempt.
Cause
Cheap unbranded motors and ESCs spun at different speeds from the same signal.
How we found it
Debugging confirmed correct PWM output. Magnets and Hall sensors on the motors then measured each motor's RPM.
Fix
We replaced them with branded motors and ESCs, which gave matching speeds and a level liftoff.
Lesson
Measure the physical output before spending more time on code.
Go deeper
08I2C slowing the control loop
The fast chip spent its time waiting.
I2C slowing the control loop
The fast chip spent its time waiting.Symptom
Timing analysis showed the loop was limited even with a fast processor.
Cause
Reading the IMU over I2C, typically limited to 400 kHz, left the CPU waiting on the bus.
How we found it
We measured control loop timing and traced where the time was going.
Fix
We moved the MPU6500 to SPI and refactored the driver, which removed the bottleneck.
Lesson
The slowest link sets your speed, however fast the processor is.
Go deeper
09Holding a hover by hand
Stable on liftoff, but not controllable in the air.
Holding a hover by hand
Stable on liftoff, but not controllable in the air.Symptom
After stable vertical liftoff, the pilot could not hold a sustained hover.
Cause
Thumb joysticks lack fine travel and spring precision, while the drone responds very quickly.
How we found it
Post-flight review pointed to over-correction by the pilot, known as pilot-induced oscillation.
Fix
Still open. Our next step is Hall-effect RC gimbals, followed by a repeat hover test.
Lesson
Your control input is part of the control system too.
Go deeper
10Radio range of about 100 m
The radio set the working area.
Radio range of about 100 m
The radio set the working area.Symptom
Beyond about 100 m in line of sight, signal strength dropped quickly and packets were lost.
Cause
The physical limits of 2.4 GHz nRF24L01+ modules, plus local interference.
How we found it
Field tests of the link at increasing distance.
Fix
Still open. We lowered the data rate to 250 kbps to get the best range. Sub-GHz LoRa or ExpressLRS is the planned upgrade.
Lesson
Design missions around the range you have measured, not the range on the box.
Go deeper
11Rough metal location
It can find a zone, not a point.
Rough metal location
It can find a zone, not a point.Symptom
Survey maps show where a disturbance is roughly, but not its exact position, depth or shape.
Cause
A consumer magnetometer plus GPS with about 2 to 5 m of error.
How we found it
We analysed the collected survey data against what we were trying to locate.
Fix
Accepted for now. We class it as a preliminary sweeping platform that guides follow-up work on the ground.
Lesson
Say clearly what your sensor can tell you, and what it cannot.
Go deeper
12Only 5 to 7 minutes of flight
High current draw limits survey time.
Only 5 to 7 minutes of flight
High current draw limits survey time.Symptom
Safe flight time is about 5 to 7 minutes, too short for wide-area surveys.
Cause
The current drawn by the motors and electronics from one 3S 5200 mAh battery.
How we found it
Flight endurance analysis of the prototype.
Fix
Still open. Battery swaps for now. A propulsion redesign is on the roadmap.
Lesson
Do the flight time sums early, and include everything on board.
Go deeper
13Little spare lift
Adding parts used up the margin.
Little spare lift
Adding parts used up the margin.Symptom
The custom PCB, modules and printed enclosures pushed the platform close to its lifting limit.
Cause
The finite thrust of an F450 with 10-inch propellers.
How we found it
Reviewing the weight added during integration.
Fix
Still open. Heavier sensors will need lower KV motors, larger propellers and a higher-voltage battery.
Lesson
Keep a weight budget from day one. Every gram adds up.