Automatic Bullseye, MOVING Dartboard

- I suck at darts, but I'm good at engineering,

which means, (board dragging)

I'm actually really good at darts.

(upbeat music)

(gleefully chuckling)

Most of the projects and builds from my YouTube channel

take a month or two to pull off.

But I've been working on this beast

with my former NASA coworker, John,

for over three freaking years, and here she finally is.

So it's fairly self-explanatory, but you throw a dart,

and as long as your initial throw

is somewhere within this diameter or so,

the dartboard helps out a little bit.

(board dragging)

(bell ringing)

And what's cool is if you use this dart,

it will make you the world's best dart player

because you get a bullseye nearly every time,

and even if you don't, it's really close.

But if you use this dart,

it makes you the world's worst dart player,

'cause it calculates the initial trajectory

and then moves the board in the exact opposite direction.

(board dragging)

(pool balls clanking, background chatter)

So eventually we took this thing to a bar

to see if it would work in the wild.

But before we get to that,

let's talk about how it actually works.

(group cheering)

Fundamentally, there are two main parts to this system.

First, you need to somehow predict where a dart

is going to land with a typical throw

from regulation distance.

And you have about 200 milliseconds to do that.

And then second, you need to somehow move the board

to that predicted location.

And you have about another 200 milliseconds to do that.

So let's start with the first part

about how we predict where the dart's going to land.

So the real secret here is that our dart

has these tiny retro reflectors on it

and then we have a Vicon motion capture system.

As humans, we can see in 3D because we have two eyes

and our brain calculates the difference

between the two images and then tells us

how far or close things are.

This is why if you close one eye,

you instantly lose that depth perception.

So that's basically what we're doing here,

except we have six eyes all around the room.

And each of these eyes or cameras

can take a 4K resolution picture 260 times per second.

But since it's hard to visually separate something so small

moving so fast in front of a busy background,

they actually use IR so they blast out IR

from all these IR LEDs and then it hits the retro reflectors

and then bounces back to the lens.

So when I wave the dart around like this,

this is what you and I see, but the motion capture system

just sees a solitary dart floating around.

One of the reasons this board took us so long to build

was that for the first two years,

we were trying to make our own motion capture system.

So we had our own cameras

and one of these awesome NVIDIA TX1 boards

which is basically like a raspberry pie on Russian steroids.

But it turns out, it's a super complex problem to solve

and the Vicon system was just turnkey.

So now that we have the XYZ position of the dart

as it travels through the air,

we use some MATLAB code to then predict where it will land.

And the trick here is that anything

you throw into the air, neglecting air resistance,

will travel in a perfect parabola.

Even something you wouldn't think of,

like a high jumper, if you track their center of mass,

it actually follows a perfect parabola shape.

So we used our understanding of parabolas

and from a side view, we used that to predict

the final up and down position of the board.

And then to predict the final

left right position of the board, we know from a bottom view

a dart will follow a linear path.

So we use good old Y equals MX plus B,

and the Y intercept, or B in this case,

tells you how far to move it left and right

for the final position.

(upbeat music)

(dart clicks, laughing) Nice throw.

So now that we know exactly where to move,

let's talk about how we actually do it.

So the board itself is on two linear sliders,

which allows it to translate to any specific location.

So that provides the track on where to move,

but the engine would be these six separate motors

you can see from the front.

Each motor has a spool and then some fishing line,

and they all attach to the center

of the back of the board right here.

Once we have the predicted bullseye spot,

the computer does the trigonometry

and then sends the commands over here to the motors.

So to move to the left, it would say like,

"You need to spin up but you need to unspin

at the same rate."

Now on the back you can see the computer signal

comes into this Arduino compatible board here,

and then you got some pre-amps here

that bump up the voltage signal

to the six individual stepper motor controllers.

And these stepper motor drivers take the moving instructions

and combine it with the power from the plug in the wall,

after it goes through this ACDC power converter.

And then finally, a bunch of voltage pulses

from the motor controller cause the motor

to take small steps and either wind up or wind down.

Now keep in mind all of this happens

in less than half a second,

and allows us to hit our sub-millimeter precision

in our board positioning.

Oh, and we don't just make one guess

on the final position of the board.

We update and refine that guess anywhere

from ten to a hundred times, which is why sometimes

you sort of see the board jitter into the final position.

(board dragging)

(upbeat music, pool balls clanking)

So we took it to a bar to see what people thought.

(group cheering)

(group yelling)

(group cheering)

In general, we found if you're really drunk,

you get so pumped at your new abilities,

if not slightly confused.

- I used to suck at darts.

(group laughing)