Precision assembly's problems shouldn't really be blamed on the robotic arm

This morning I saw someone in the group forwarding that news about Xiaoyu Zhizao's precision assembly, and the discussion underneath was basically all about the repeat positioning accuracy of the robotic arm. My feeling after reading it was that the direction might be a bit off.

Precision assembly, from the front end, looks like just an assembly motion. For standard tasks like hole-shaft fitting and flat surface press-fitting, today's force-controlled robots generally come with preset process packages — the operator just configures a few parameters and it runs, no need to write force-control algorithms yourself. This step really isn't complicated. I asked a friend who does integration, and he said a single station can be debugged and running in a few days.

The trouble is further down the line.

Core alignment, deviation judgment, whether the fixture is stable, how to inspect after assembly — these things are bundled together. The time saved at the front end often gets paid back with interest later. I once worked with a small-batch production line where the robotic arm itself was fine; the sticking point was that after the fixture was changed over, the datum drifted, and the deviation judgment for a whole batch went haywire along with it. In the end what got reworked was the inspection logic, not the robot. That batch dragged on for nearly two weeks, and the problem wasn't in the hardware.

This is the same class of problem as the power wall we hit on chips. No matter how good a single module's specs look, if the system boundary can't be pinned down, the whole thing is unstable. You can't optimize just one point — you have to draw the boundary first before talking about optimization.

So now when someone asks me whether this stuff is worth adopting, I'll first ask back: have the fixture and inspection plan been decided? Once they're decided, then talk about the robot; if not, don't even start. If this order is reversed, everything afterward is just patches.

Industrial robots have been in a fierce spec war these past few years, but the place where the real gap opens up has never been on the spec sheet.