
A 6-Axis Force Sensor on the Wrist Still Can't Handle Fingertip Tasks
I spent the weekend tinkering with this LB200 and hit quite a few pitfalls. Strictly speaking I didn't buy it—I followed along for half a day while a customer installed it, then shamelessly borrowed it and hooked it up to my own small arm to try for an evening.
First, let me be clear about what this thing is. A six-axis force sensor measures force in three directions (Fx, Fy, Fz) plus torque in three directions (Mx, My, Mz), outputting six signals together. The robotic arm itself only knows its own joint angles; it doesn't know what the end is touching or how hard. Adding this gives it a sense of touch. The LB200 is from Bluepoint Touch, positioned for the wrist or base of collaborative arms and heavy equipment, not a fingertip component.
The first pitfall in the prep stage is the mounting surface. The instructions emphasize that the mounting surface must be flat and the bolts tightened evenly in diagonal order. The first time, to save time, I tightened the four screws going around in a circle, and the zero point was off as soon as it powered on—I didn't note exactly how much, but before a finger even touched it, the reading was already off zero. I took it off and tightened diagonally in two passes, and then it was fine. This thing is sensitive to assembly stress; mount it even slightly crooked mechanically and software can't compensate later.
After powering on and watching the data scroll, this part was cleaner than I expected. Sitting unloaded, all six channels jump slightly around zero; push from the side with a finger and Fy clearly changes while Fz barely moves; press down and it's the reverse. This is called crosstalk—the false reading a force in one direction causes on other channels. On my side the crosstalk didn't look large, but I didn't quantify it, so I won't draw conclusions.
The most time-consuming part was zero calibration and gravity compensation. I had a dexterous hand mounted on the end, and the hand itself has weight that's constantly pressing on the sensor. Zero it lying flat, and the readings all change once you flip the pose. Either fit the weight and center of mass of the end load for compensation, or re-zero at every pose—the latter is completely unusable on a real machine. I burned most of the evening on this. So my judgment is direct: the hardware barrier for six-axis force isn't high; what's hard is the calibration and compensation software workflow after installation. Whoever makes that workflow smooth wins.
Another detail is cabling. The force of the cable being pulled goes straight into the signal; I fixed the cable to the arm with zip ties and left a bit of slack, and after that it was stable. Also, the zero point shifts a little right after power-on versus after running for half an hour, so you need to warm up—for precision assembly this can't be ignored.
I didn't hit much resistance on the interface and configuration side; zero calibration and filter parameters are all configurable, and hooking into my own acquisition script went smoothly. I looked at its data alongside the tactile array on the back of the hand, which was interesting: the wrist six-axis sees overall contact force and torque, while the fingertip array sees contact surface distribution—the two are complementary, neither can replace the other.
While searching for material I also came across a miniature six-axis with a 9.6mm outer diameter and 3 grams—that's the fingertip route, a completely different species from the LB200, don't mix them up.
Conclusion: it depends, but I lean toward recommending it. Suitable for teams doing collaborative arm drag teaching, force-controlled grinding and assembly, and needing overload protection—especially if the arm side already has a gravity compensation framework, it plugs in quickly. Not suitable for people who want to use it to replace fingertip touch—a single wrist six-axis can't tell where the contact point is; also not suitable for those who only want collision detection—current loop plus threshold is enough, no need for this. And calibration: the user end can only do zeroing; the real sensitivity calibration has to go back to the factory. Several domestic vendors emphasize their own calibration lines—that's genuinely a barrier, but it also means if something goes wrong you have to ship it back.
Looking ahead, I think six-axis force will converge in two directions. One is going into the joints, as integrated joints with built-in torque sensing, which will eat away part of the standalone wrist unit; the other is shipping with temperature compensation and load adaptation, taking the gravity compensation work out of users' hands. Right now the most blocking step is that one evening of calibration and compensation after installation.
Physix Frontier