
Fitting dual same-side encoders into a joint: wiring, not precision, is what's blocking me
I compared Gongwang's ultra-narrow-frame same-side dual encoder with another company's DPB series, and actually ran it. But I should be clear upfront: I can't take an encoder home to test — I ran this at a customer site following an installation, spending most of a day there.
Here's the situation. The customer is building integrated joint modules, previously using a single-encoder setup that only measured the motor side. The problem was the reducer — harmonic reducers have backlash, so when the motor reaches position, the output end's actual angle is still a bit off. For position control it barely works, but once you want collision detection it doesn't.
The dual-encoder idea isn't complicated. Put a read head on each of the output side and the motor side; when the output end receives external force, the output side's angle changes first, while the motor side hasn't reacted yet due to the reduction ratio — subtract the two and the difference is the estimate of external torque. I'd read a post explaining this before and understood the principle, but actually installing it into a joint is another matter.
Gongwang's unit's form factor is ultra-narrow frame plus same-side wiring. Narrow frame, as I understand it, saves radial space — with the same outer diameter you can keep a larger hollow bore, making wiring easier. Same-side means both read heads are on the same side, with the harness exiting from one side.
| What I cared about | What I confirmed on site |
|---|---|
| Whether dual readings are same-side | Yes, both read heads exit on one side |
| Output resolution | Dual 20-bit absolute, position known on power-up, no homing needed |
| Frame width | A ring narrower than common models, hollow bore can indeed be larger |
| Air gap tolerance | Assembly instructions require it fairly strictly; I didn't actually measure boundary values |
| Force estimation | Encoder only gives two angles; the difference algorithm is on you |
The install was more of a hassle than I expected. The radial space saved by the narrow frame got eaten right back by the harness. With same-side wiring in a joint with short axial length, the harness bend radius wasn't enough — the first install pressed against the read head. We had to modify the wiring channel before it passed.
Another pitfall is magnetic ring eccentricity. The assembly instructions specify air gap requirements but don't say how to control eccentricity. We checked with a gauge — once eccentricity exceeded the limit the waveform was visibly uneven, and a periodic fluctuation showed up in the dual-encoder difference. We ended up solving this with a fixture, not the encoder's fault, but the docs should mention it.
What surprised me was the absolute value. With the old single-encoder setup, every power-up required homing — on the production line you'd crank each one by hand, labor-intensive. This one reads position directly on power-up, saving that step.
Another value I didn't expect is explainability. I did AI underwriting for a few years, and I habitually ask "after something goes wrong, can you explain why it happened at the time." The dual-encoder difference sequence is itself an evidence chain — if it bumped or jammed, you can look back at the logs and see how the external torque climbed. For teams putting collaborative robots onto production lines, this is worth more than precision specs.
Conclusion: it depends. Suitable for teams already building integrated joint modules, with tight axial space, needing collision detection — especially scenarios requiring after-the-fact traceability. Not suitable for two kinds of people: one, those who want to buy one and solder a board themselves to play with — the supporting fixtures and calibration aren't something an individual can put together; two, teams without calibration capability — the encoder only gives you two angles, and the force estimation thresholds and filtering are all on you to tune; tune it badly and you get false stops every day.
An action suggestion. If you have an existing joint structure on hand, first install one and measure three things: whether the harness bend radius is enough, whether magnetic ring eccentricity is controllable, and how big the noise floor of the dual-encoder difference is. That third number determines how low you can push the collision threshold — without that number, it doesn't matter how many bits the spec sheet says.
Physix Frontier