At @BuzzRobotics, we 3D scan our customer (one time task) and then use that to do path planning on our robot. Here's a path from the middle of the neck to the middle of skull. What you're seeing is fully dynamic - no hard coding involved.

Here's how we detect pressure over at @BuzzRobotics. Included is a demo of cutting my hair. PS - I'm running out of hair on my body because of test runs. Send help.

At idle, readout from our force torque sensor post low pass filter. Top is with the clipper off, bottom with it on. Left axis is newtons, bottom is # of samples (100hz sampling rate so divide by that for seconds)

For sensing pressure, our @BuzzRobotics robots use a force torque sensor. They tend to be noisy so we're using a low pass filter to clean the signal up. Top is original signal (idle, pull, then push). Middle is exponential moving average filter. Bottom is butter filter.

Over at @BuzzRobotics, we train our robots to understand skull size and curvature using a computer vision system we built that (1) scans our customer via @scandyco and (2) computes the normal vectors along a curve between two points of our choosing.

Using a 3D scan, here's a path we generated from the neck to the crown of a real head (mine!). Normals are purple. More to come.

We bought a turntable off Alibaba and hacked it. It's where our @BuzzRobotics customers will sit.

Hooking up our human grade turntable (~330lb load) to a usb relay. We'll be able to control turn direction as well as speed (the built in controller senses for a simple falling edge signal). Next up will be attaching a stool to it.

Our customer will sit on a rotating platform and they'll have some stickers on them that look like QR codes. Our robot will need to rotate them different amounts, based on which part of the haircut we're on. We do that using a USB relay, some computer vision, and a turntable.

One challenge we face at @BuzzRobotics is identifying (1) which part of the human head we're looking at and (2) how to orient our clipper so that it is level (x/y/z) and has the correct angle of attack (rx,ry,rz). ArUco markers help us do both.

Many of you have reached out to us at @buzzrobotics during this crisis. Due to your feedback, @P_deeeee and I have now accelerated plans to deliver a full buzz cut. Mannequin trials are underway and human trials will start next week.

Last Thursday, @BuzzRobotics delivered the first ever robotic haircut in history to @kwilliamshaw, for $6. Before / after: i.imgur.com/JDkPlGB.jpg Receipt: i.imgur.com/pGharrs.png With COVID-19 here, we need to rethink what roles robots play. Let's start that discussion now.

Live trials for @buzzrobotics neck fuzz cut on me. Left is before, right is after.

Final trials (2x speed video)

Today, @BuzzRobotics enters open beta. We're charging $6 for a neck fuzz cut (tip included). Want to be among the first people in the world to get a robotic haircut? Shoot me a DM. Video of our entire process 👇

Trying to 3D scan your own neck is hard. A robot arm + velcro helps, though.

Here's a full cut on my forearm in our final trials before servicing neck fuzz cuts. Also a preview of the kiosk setup: status screen up front, chair in middle, bot behind you. (video is 2x speed)

Normal angles are in place; everything you see is being dynamically computed except manual alignment with the starting point. Jerky transitions between traces could be smoother though. Don't mind my hand - the head & tripod are unweighted. (2x speed video)

Here, we're building a bridge between the physical world and the digital world. Our bot now knows how to take a point on our 3D scan and orient itself in preparation for tracing/cutting a vertical path on top of it.

Before we perform robotic haircuts at @BuzzRobotics, we have to add some constraints to the experience. One constraint will be around movement; we want our customers to stay as still as they can. Here's our computer vision work that performs that. More in thread 👇 around how.