Planetary Integrated Joint Modules: Starting with the Low-Torque Set

I spent two days trying Quanzhibo's PA series planetary integrated joint modules. Let me first clarify what I got my hands on: the lab received two sets of low-torque ones last month, the 2Nm range, and I ran them on the test bench for two days. The 400Nm ones I can't lift, I only saw a sales demo at a customer site, so the high-torque part is an inspection, not a hands-on test.

The term 'joint module' confuses first-time hearers. You can think of it as a robot's shoulder or knee: a shell containing a motor, reducer, encoder, and driver, with only power and communication wires outside. Previously, building a robotic arm required assembling these yourself, then wiring and tuning parameters one by one. The PA series uses planetary reduction, i.e., gear meshing, which compared to harmonic reduction has lower precision but is cheaper, more robust, and impact-resistant.

The preparation phase took more time than I thought. It's a standardized matrix from 2Nm to 400Nm, which sounds like you just select by torque, but in practice you have to distinguish peak torque and continuous torque. I initially only looked at peak, and the selected model started overheating after running continuously for over ten minutes on the bench. Later I switched to leaving margin based on continuous torque, and the problem disappeared. This step is easy to mess up without someone reminding you.

The first step is wiring. Power lines plus a CAN bus—CAN is the most common communication line in robotics, two differential wires transmitting data, good anti-interference, and you can string several joints together. After wiring and powering up, I hit the first pit: I gave power first, and the driver directly errored and wouldn't start. The correct order is control power first, then power; when shutting down, reverse. The manual says this, I didn't read carefully.

The second pit is zeroing. These modules generally have dual encoders: one at the back of the motor, one on the output flange. The motor-end one measures rotor position, the output-end one measures the actual joint angle. The output-end reading is not calibrated at the factory; the first time I read it, the joint angle kept drifting, and I thought I wired it wrong, wasting a lot of time before realizing I needed to do a zero calibration first. The calibration process isn't hard—just a few clicks on the host computer—but it's really frustrating when you don't know. Last month I wrote about how to get started with edge AI chips, and mentioned that hardware pitfalls are basically in power-up sequence and calibration, and this proved it again.

The pleasant surprise is the level of integration. Our previously self-assembled joints took half a day just for wiring, and the wires easily wore out during rotation. With the PA set, after connecting two wires it moves; from unboxing to getting it to rotate took about 40 minutes. Also, the planetary solution's impact resistance is indeed good—I accidentally hit the limit on the bench, and there was no problem. With harmonic, that would have made my heart skip a beat.

What I'm not satisfied with: planetary reduction inherently has backlash, that small gap between gear teeth. You can't feel it in position control, but if you do force control or manually push the joint in reverse, you can clearly feel that dead zone. While searching for information, I came across an article on integrated joint design that classified actuators into rigid, series elastic, and quasi-direct drive based on sensor schemes. Series elastic means inserting an elastic element in the joint to measure torque; quasi-direct drive is low reduction ratio with a high-torque motor, estimating torque from current. The PA set leans toward the quasi-direct drive path, with force control not relying on elastic elements. So for scenarios requiring sub-millimeter repeatability or precision force control, you still need harmonic.

My conclusion is it depends. For robot prototypes, teaching benches, teams needing to quickly build joints without being tormented by wiring, the PA set is very convenient; for extreme precision and precision force control, don't choose planetary.

Looking ahead, I think planetary solutions will take a significant share of the mid-to-low-end joint module market. The legs of humanoid robots and below the elbow of industrial arms don't need harmonic precision anyway, but they need to withstand falls and impacts.