No matter how good the tactile sensor, calibration is still unavoidable

Let me say something. These past few days I've been trying out Daimon's DM-Tac visuotactile sensors. I looked at the specs for the W2, X, F, and G models, and my most direct impression is that this thing really is strong at capturing tactile information — it can pick up force, texture, and temperature. They say it's the world's first multi-dimensional high-resolution visuotactile sensor, using a monochromatic light principle. When grasping, the tactile information is far more stable than relying on vision alone.

But that's not what I want to talk about.

I tried it against that batch of cardboard boxes and stretch film in the warehouse, and the feel really is richer than ordinary solutions. The problem is, once it's mounted on the robotic arm, that calibration workflow still hasn't gone away. No matter how accurate the sensor itself is, once it's installed you still have to re-align the position, the contact surface, and the force zero point. On my end, just the alignment step alone took me half a day, and if an ordinary camera jitters you have to start over — this is the same pitfall I ran into before with static guide plates. The finer the sensor makes the tactile sensing, the higher the requirements for upfront alignment; once error creeps in, all those high-resolution advantages get discounted.

Daimon released the DM-Tac W in May 2025, and around the same time also paired it with the five-finger dexterous hand DM-Hand1.

The technical foundation of this sensor series is the real deal, not just a PPT. But between the lab and the warehouse production line there's a whole set of calibration and integration work. Whoever smooths out that stretch is the one who can actually put it to use. At this stage, I'd suggest trying it out at one or two fixed workstations first — don't roll it out everywhere right away.