For Beginners Installing 6-Axis Force Sensors, Zero Calibration Is Usually the Sticking Point
Community Discussion · Humanoid Robots

For Beginners Installing 6-Axis Force Sensors, Zero Calibration Is Usually the Sticking Point

Whole Machine TheoryWhole Machine Theory1d ago2026/11/17 116 views

I compared a six-axis force sensor that outputs the raw bridge signal against one like the LA45 that does onboard decoupling and directly outputs force/torque values, and actually ran both. Here's the conclusion up front: if you're a beginner trying to get force control working once, pick the latter—it saves you a lot of debugging time. The former leaves the solving work to you; you have to write your own algorithm on the host computer and also calibrate crosstalk. During the prototype stage, there's no need to burn time on that.

First, two terms. A six-axis force sensor measures thrust in three directions and torque in three directions simultaneously; mounted at the end of a robotic arm, it's like giving the robot's wrist an organ that can feel how hard it's being pushed and how much it's being twisted. Onboard decoupling means the sensor itself computes the raw signal into force and torque and gives you usable numbers directly, so you don't have to solve it again.

According to the documentation, this type of product outputs solved force and torque, not the raw bridge signal.

Below is my whole process from unboxing to getting drag teaching working. Follow along and it'll take about an afternoon.

Mount it right first, then talk about reading accurately

1. Measure the hole pattern. Note down the hole spacing and bolt specs of the robot arm's end flange and check them against the LA45's mounting holes. Don't skip this step—I've seen people get halfway through mounting before realizing the holes don't line up and have to go back and remake the adapter plate.

2. When mounting, check the force direction marking. There's an arrow or engraved line on the sensor pointing toward the end effector side. Mount it backwards and all the force value signs will be reversed later, and you'll doubt your whole life while troubleshooting.

3. Leave slack in the cabling. The cable moves with the arm, so don't cinch the zip ties tight—leave some play, or the core will break after a few runs.

4. Connect to the host computer. On my side I use a USB adapter; after plugging in, you can see the serial port in Device Manager. Install the driver, open the bundled software, select the right serial port and baud rate, and click Connect.

The expected result is that the six channels on the interface start showing jumping numbers; at rest they're near zero but not exactly zero, which is normal.

Zeroing is the step most likely to go wrong

5. Zero under no load. Don't mount the tool on the end yet; position the arm horizontal and stationary, and click Zero Calibration in the software. Wait two or three seconds, and the six channels settle near zero.

6. Mount the end tool and read again. At this point the force channels will have a fixed offset because the tool's own weight is pressing on the sensor. The software usually has Load Compensation—enter the tool weight, or click calibration again to subtract this part.

7. Push the end by hand and watch the values change. Push in one direction and the corresponding force channel increases; release and it returns to zero. Don't just push in one direction—try sideways, vertically, and twisting, and confirm all six channels respond.

By this step, you've got a wrist that can read force.

There are a few pitfalls. Mounting the tool without doing no-load zeroing means the zero point carries the tool's weight, and all the force protection thresholds will be wrong afterward. An uneven mounting surface or uneven preload on the four bolts causes crosstalk—you're clearly only pressing down, but the lateral channels jump too. Take it off and check the mounting surface for burrs, and tighten the bolts diagonally in alternation.

If the baud rate or sampling rate is set too high, the data starts jumping; drop it one notch and it stabilizes. If the drag threshold is set too high you can't push it; too low and it triggers at the slightest touch. Adjust from small to large and find the value that just barely lets you drag without false triggers.

Once it works, I suggest trying over-force protection next: set a force limit, and if exceeded, have the arm stop. This scenario is closer to production line acceptance than drag teaching—the first thing customers ask is usually "will it stop if it crashes." Once this runs smoothly, move on to constant-force contact and precision assembly.

Don't chase precision specs right out of the gate. Get the three things right first—zero point, direction, threshold—and you're halfway down the force control path.

2 replies

?
Ctrl + Enter to reply
Production Line Old Li

We tried force-protection e-stops on our line. After a changeover and reinstalling the tool, if you don't re-zero, all the thresholds drift and yield drops two points.

Mo Mo
Mo Mo22h ago

"Installed backwards so all the force value signs are flipped" — that's so real. The first time I tuned force protection the sign was reversed, and it kept loading even after a collision.