2012 · Physical AI before it had a name
Robo-Vision — Physical AI before it had a name
An automatic meter-calibration robot that sees a printed dial and pointer through a USB camera, decides with fuzzy logic and acts by adjusting magnetic strength, in a closed loop.
- Years
- 2012–2015
- Organisation
- Alsons Auto Parts (Alsons Group)
- Role
- Senior Research Executive (R&D lead)
- Input
- Processing step
- Output
Context
Meters have to be calibrated so that the pointer lands on the right mark. At Alsons Auto Parts I led R&D for process improvement, and Robo-Vision was the result of extensive work on that problem: an automatic calibrator robot for meters.
What I built
A machine that closes the loop between seeing, deciding and acting, without a person in it.
- Perception. Live image analysis from a USB camera detects the printed dial and the pointer.
- Decision. Fuzzy logic minimises the error between the calibration point and the pointer position.
- Actuation. Calibration is made precise by reducing magnetic strength: high-voltage driver circuitry and a microcontroller create an alternating magnetic pole.
- Software. A Windows application with a C#.NET front end and C++.NET back-end processing using native calls.
How it works
The camera reads where the pointer is. The fuzzy controller compares that with where it should be and decides how much to adjust. The driver circuit applies the correction magnetically, the meter responds, and the camera reads it again. The loop repeats until the pointer sits on the calibration point.
It is the same sense, decide, act structure that Physical AI systems use today, built in 2012 with classical vision and control.
Stack
C#.NET and C++.NET on Windows, USB camera image analysis, fuzzy logic, a microcontroller and high-voltage driver circuitry.