Gripper keeps crushing thin-shell parts—how to tune it with tactile sensing

I spent two days trying out Modulus Tech's industrial gripper tactile sensing, running a batch of mixed-material workpieces through it. Conclusion first: this thing doesn't make you smart just by installing it — calibration and thresholding are two steps you can't skip.

Background first. The customer has one gripper that needs to grab three types of material in rotation: metal parts, plastic parts, and a thin-walled shell part. Metal parts can take a squeeze; the thin shell dents if you clamp too hard. The old approach was to store a separate parameter set for each material, stop the line and re-tune on changeover, and rely on a veteran worker listening to the sound and watching for deformation to dial it in. Slow, and it depends on who's on shift.

Tactile sensing sounds fancy, but put simply it's a layer of electronic skin stuck on the gripper fingertips that can feel two things: how much force is being applied, and whether it's slipping. The gripper is those two fingers at the end of the robot arm, responsible for pinching and lifting the workpiece. Force feedback sends the force felt at the fingertips back to the controller, letting it decide on its own whether to tighten a bit more. Modulus's setup is integrated into the gripper, not an add-on module you wire up yourself.

Here's my process from scratch.

1. First mount the gripper on the end of the robot arm, connect the comms cable and power. Mine is a standard end flange, just line up the holes and screw it on. After mounting, manually flex the two fingers to confirm there's no sticking, then check whether the gripper status light on the controller is solid. Solid means the power-on self-test passed; flashing red is most likely a loose cable.

2. Open the companion debugging software, create a new grasping task. Select the gripper model, give the task a name, say "mixed-material test grab." Two force curves will appear in the interface, corresponding to the left and right fingertips. Without tactile sensing installed this is a flat line; with it installed, it jumps as the fingers open and close.

3. Do the calibration. Don't skip this step. Calibration tells the system what "empty hand, nothing touched" looks like. Open the gripper fully, click zero calibration, wait two or three seconds, and the curve settles near zero. Then take a standard block of known weight, clamp it once, and let the system learn how much force corresponds to how much force value.

4. Set the threshold. The threshold is a boundary value; exceed it and the system considers it "clamping too hard." For thin-walled parts I set it at a lower level, for metal parts higher. In the software you drag a slider or just type the value. Save it as a recipe, bound to the workpiece ID.

5. Run the test grab. Do an empty run first to check the open/close sequence is right. Then load real workpieces, grab them one by one. In my testing, after about twenty-some pieces, the thin shell parts stopped getting crushed. Once in the middle it didn't grip firmly, the fingertip detected slight slipping, and it added a bit of force on its own and picked it up.

The pitfalls are mainly in three places. First, setting the threshold without calibrating — the curve baseline drifts, so no matter what you set it's off. Second, poor cable routing — when the gripper opens and closes it rubs the cable, the force curve suddenly jumps, and it false-reports as slipping. Third, the threshold set too tight — the slightest vibration triggers an alarm and the line stops constantly. What I ended up doing was leaving a bit of margin on the threshold, splitting alarms into two levels, and having slight slipping just add force without alarming.

Comparing the two approaches:

Item	Manual parameter storage	Tactile feedback
Changeover tuning	Re-tune for each material	Load recipe, fine-tune threshold
Crush risk	Depends on experience and trial and error	Force hits the cap, backs off automatically
Dependence on experience	High, blind without the veteran	Medium, just need to read the curve
Tuning time	In my testing, fairly long	First calibration takes time, saves after

That's the effect, and I'll spell out the downsides too. Tactile sensing solves "is the grip stable," not "where to grip." If the workpiece position is off, it still grabs air. Also the sensor itself is a bit delicate — one bump and it can drift, so you need periodic recalibration.

Having learned this, what to try next. You can try combining tactile and vision. Vision first finds where the workpiece is and roughly what shape, tactile confirms after contact whether the grip point is right, and if not, release and try another point. Several papers I've read are already working in this direction — the sensor fingertip provides extra information to the planner, helping it find the right grasp point. In logistics scenarios where boxes are piled mixed together, this combination is more reliable than vision alone.

On the trend, I'll make one bet: in the next two years, tactile sensing will go from "optional" to a default part of the gripper, just like force control did back then. The reason is that line changeovers are getting more and more frequent, and the manual parameter-tuning model can't keep up. What really bottlenecks isn't the sensor itself, it's the software workflow of calibration and recipe management — whoever smooths that part out is the one who can actually sell.