From a robot leg to the factory floor

My final-year project was a 5-DOF humanoid robot leg — torque distribution, multi-axis sensor feedback, dynamic stability. It won 2nd place at the STEM All Pakistan Project Competition, and I thought it was the hardest thing I'd ever build.

Then I joined a factory.

The lab lies to you, gently

In the lab, your camera is clean, your lighting is constant and your power never flickers. On a production line, cameras collect dust by Tuesday, sunlight moves across the floor all day, and the network drops exactly when the shift supervisor is watching.

The vision inspection system I deployed works not because the model is clever — it works because of everything around the model: the RTSP watchdogs that restart dead streams, the failover logic, the reject log that writes to SQL even when the dashboard is down.

What transfers

Control theory transfers. The humility of tuning a PID loop — change one thing, measure, repeat — is exactly how you tune a detection threshold on the line. And the robot leg's biggest lesson transfers too: the system is never just the algorithm. It's the sensors, the mounting, the cables, and the person on the night shift who has to trust it.

If you're a student: build something physical that has to work for more than a demo. That's the whole job.