Ball Balancing
A practical study in feedback control: sense, decide, actuate, stabilise.
01overview
A control-systems study in keeping a ball on target by continuously sensing position, computing error and driving actuators to correct it — the classic feedback loop made physical.
The project combines a PID controller with a camera-based perception pipeline (an EfficientDet-Lite ball detection model) and actuator/servo drive code, plus GUI and web interfaces for observing and tuning the loop.
It exists to develop practical intuition for stability, gains and tuning in real time. Recorded experiment videos document the system in action.
02architecture
- 01▣
Perception
camera · TFLite detection
- 02▣
Estimate
ball position
- 03▣
Control
PID · tuned gains
- 04▣
Actuation
motors · servos
- 05▣
Plant
platform · ball
03stack
05software
- Python control software (module suite)
- PID controller
- Camera-based ball detection (TFLite)
- GUI + web interface (prototype)
- Robot kinematics module
06problems & solutions
Latency between perception, control and actuation breaking the loop.
Keeping detection lightweight (on-device TFLite model) and closing the control loop as tight as possible. Tuning data is being logged for publication.
PID tuning on a physically responsive system.
Iterative gain experiments on the live rig; videos of successive runs are being collected as evidence.
07results & specs
- Perception
- Camera + EfficientDet-Lite (TFLite)
- Control
- PID implementation
- Interfaces
- GUI + web (prototype)
- Stability metrics
- Documentation in progress.
08gallery & media