A week following general-purpose robots: the good and the bad

Last month a customer's production line got a general-purpose robot, from Shenshi Weicheng, and they asked me to take a look. I don't usually touch industrial robot hardware — that's the integrator's job. But the word "general-purpose" made me curious. Everyone calls their product general-purpose; how general is it really?

Day one was basically installation. Base, body, control cabinet, and teach pendant were packed separately, and it was already 8:30 by the time it reached the factory. The first thing after powering up was finding the zero point — letting the robot know where its origin is. This step was slower than I expected. The teach pendant interface is that very industrial tree-menu style; finding the calibration item took digging through three or four levels. The engineer said they usually use another brand, and this logic takes half a day to get used to. After calibration I tried dragging the body by hand. The damping on the six joints was tuned decently — I could drag it with one hand, which is more convincing than what's written on the spec sheet.

Day three we started running real tasks: grabbing workpieces from a bin and placing them into a positioning fixture on a rail. It got stuck once midway — the grasp point was off by about two millimeters, the workpiece jammed when placed, and the line stopped for over ten minutes. Re-teaching one point fixed it. Teaching, by the way, means physically guiding the robot through a position by hand and having it record it. In the afternoon there was a pleasant surprise: they switched to another workpiece of similar size, only changed the gripper and modified two points, and it was running in twenty minutes.

A week later I went again and asked the line supervisor how it actually felt. He said stability hadn't been a problem, no dropped balls during continuous runs, but changeovers still required re-teaching. There's an article on Huxiu that talked about this:

The closer you get to task understanding, the easier capability is to share. The closer you get to physical contact, the harder it is to bypass differences in the body.

This time I got a feel for it. Task-layer stuff really can be reused; the grasping moment still has to be tuned point by point on site.

The downsides have to be mentioned too. Thin documentation, interfaces that aren't exactly generous, and writing your own logic on top is awkward. The software ecosystem is a notch behind the top few players — that's the truth. As for the boundary between the hardware itself and integration solutions, they're probably still figuring that out themselves.

The upsides are just as obvious. Solid build quality, decent drag feel; in high-mix low-volume scenarios, changeover time really can be compressed — the supervisor said from two days to one, though it hasn't disappeared; local service responds fast — I called and someone was on site the same day.

Conclusion: it depends. Manufacturing teams doing high-mix low-volume work with an integrator alongside them can put it on the shortlist. If you want to buy one and build it yourself, don't touch it. Pure algorithm experimentation isn't a fit either — this hardware isn't for that.

General-purpose is relative. The closer you get to physical contact, the less general-purpose it becomes.