My Dad Builds Robots That Chew Cheese

The title is the short version, and it is accurate. The longer version is that my dad retired from Colgate-Palmolive as a senior director, after years of working out why one toothpaste feels right in your hand and another does not, and started a company with an engineer named Dmitry to build machines that answer that question without asking a room full of people. The company is called Perception Instruments. I am the third member, and I am the one who does not understand the science. In late October the three of us took two of the machines to a rheology conference in Boston. This is what they are, and what I actually did.

The Spreading Tribometer in Boston. Frame from the footage we shot at the booth.

The problem is that people are the instrument

If you make toothpaste, lotion, sunscreen or cheese, the thing you are selling is partly a feeling. Is it sticky. Does it spread. Is it stringy when you pull the spoon away. Is the tube hard to squeeze.

The reliable way to measure a feeling is to ask humans. Companies run expert panels, where trained people score stickiness or spreadability on a scale of zero to ten, and home use tests, where real customers take the product home. Both work. Both are slow and expensive, so product developers would much rather put a sample in a machine and read a number.

The machines they already own measure material properties, like viscosity. My dad's point, which he makes better than I do in a five minute video, is that a property is not a perception. Ask someone how thick a liquid is and the answer changes depending on whether they stirred it with a finger or watched it pour, and it changes with the color and how clear it is. The human mind mixes things together. A viscometer does not.

So there is a third kind of test, called emulative: build a machine that does what the person does. Do not measure the cream. Measure what happens when a finger spreads the cream.

Two robots

Both machines look a bit like 3D printers, and that is a fair mental model. A head moves precisely in three axes inside a frame. Instead of a nozzle, the head holds a finger. Or a tongue. Or a set of teeth.

I drew them on my whiteboard before the conference, mostly to explain them to myself.

My pre-conference cheat sheet. The scientists have better diagrams.

The Spreading Tribometer rubs things. Tribology is the science of friction, and this machine moves a ball or a finger-shaped probe through a thin film of product in controlled circles, the way you rub lotion into your arm. While it moves it records how hard it is pressing down and how hard the product is pushing back sideways. Divide one by the other and you get friction, plotted against speed on a screen bolted to the side.

It also watches. There are three cameras inside: one for the whole chamber, one on the probe, and one looking straight down at the sample. The overhead camera takes seven photos under seven different lights and reconstructs a height map of the surface, so you can see the film the finger left behind, in 3D, from the same run that produced the force curve.

A probe spreading cream in circles. Every pass is the same pass, which is the whole point.

The Advanced Texture Analyzer presses, squeezes, pulls and chews. The ordinary texture analyzer is the workhorse of this industry, and it does one thing: it goes up and down. People do not chew up and down. The lower jaw slides sideways. So this one has a model jaw with upper and lower teeth, and the lower teeth move laterally while the upper ones come down.

It chews cheese. At the booth it also spent hours on a marshmallow, because a marshmallow can be chewed more or less forever and a demo has to run all day.

A model jaw chews a piece of cheese. Science.

The jaw comes off by hand and a finger goes on. The demonstration set has eleven attachments: fingers in several materials, one with a fingernail, a tongue, a paddle, a cluster of pins. Change the contact and you change the test.

None of this is a party trick. One example from my dad's published work: a consumer panel wearing a haptic glove showed that people squeeze a toothpaste tube with about 15 newtons of grip. Apply 15 newtons with the machine, measure how fast the paste comes out, and you have a lab number that stands in for "is this tube annoying." There are 22 peer-reviewed papers behind the methods. I have read the abstracts.

What I did

I run a software company. My job in this one is everything that is not the science or the hardware, and for the conference that meant a short list with a hard deadline.

A sign-up sheet, in thirty minutes. The day before the hall opened I had half an hour before I had to leave and nothing to put on the table. I worked with Claude to build a one-page sheet: six big rows for a name, an email and a company, a checkbox for which machine they cared about, and a QR code in the corner. We printed twenty. My contribution to the experimental design was offering to buy travel-size toothpaste at the airport so we would have something to spread.

A page for the QR code to land on. It went from nothing to live on the site that evening, with a form that fed the same lead sheet as everything else, tested end to end before the doors opened at 8:30 the next morning.

The booth. Booth 7 at the Society of Rheology's 97th Annual Meeting: eight feet by ten, one six-foot table. My dad gave a talk. Dmitry ran the machines. Rachael and I worked the table, with a line I could deliver without a doctorate: tell us the thing your customers complain about, and we can measure it.

Here is the honest result on the technology. The QR form got zero real submissions. Not one. Every lead from three days came from the clipboard and from my own notes after each conversation: 21 people, and four companies that asked for a demo. At a conference of scientists, the paper sheet beat the web form by twenty-one to nothing. I will remember that the next time I am tempted to solve a problem with a landing page.

The footage. For the camera people: everything was shot on a DJI Ronin 4D in 8K ProRes RAW, which is a ridiculous amount of camera to point at a marshmallow. The files were enormous and the standard tools could not decode them, which turned into its own article. The resolution was not for show. Those frames are detailed enough to measure from, and they are what the 3D model of the instrument on the product page was checked against, down to the proportions of the cabinet. What came out the other side is on the videos page: four short parts, about seven minutes in total.

The videos went through twelve versions, because the reviewers were the two people who built the machines and they were right every time. The cheese had to be in. The clean friction curve had to be in and the noisy one out. A caption said a camera photographs the film, and the correction came back: it is three cameras, and the top one takes seven frames. I do not get that kind of review on software.

The debrief. The three of us did a 66 minute call after the show. I had it transcribed on my laptop with the speakers separated, and turned it into a list of 19 decisions with an owner on each. That list is what we have been working from since.

The site. In the three weeks after Boston the site took 28 merged pull requests: the videos, a player, a product page where you can orbit a 3D model of the instrument, mobile fixes, and the unglamorous work of making sure email from the company's domain actually arrives.

The part that is mine

There is one more corner of the whiteboard.

The question I wrote for myself.

Where does the measurement data go? A run produces force curves, video, surface maps, temperatures, motor currents. Right now the honest answer on my whiteboard is "csv, excel, SQL," with a comma between each and no decision. That is a data problem, not a rheology problem, and it is the first thing in this company that sits squarely in what I do for a living.

I did not expect to care this much. I thought I was helping my dad with a website. What I found is a retired scientist and an engineer who built a robot finger that can tell you, with a number, why one cream feels better than another, and a jaw that chews cheese sideways because that is how people chew. It is the most interesting hardware I have been near in years, and my dad made it.

If you make something people touch, taste or squeeze, go look at the machines.