A little while ago, I ordered an IMUduino which is a tiny arduino clone that has a full IMU + bluetooth low energy built into it. This is the perfect part for my next prototype of my parkour impact sensor.

In order to tweak the led reactions, I want to be able to record and playback IMU data, so that I can not exhaust myself while tesitng out different light patterns.

So I need to build a record and playback system that allows me to visualize what happened in the recording, and play that back to the device to run through how the leds will react.

But first, I need to just familiarize myself with the device, the data it can return, and just generally figure out the system for record and playback, so I built a small system that reads data from the device, and sends it to the web client running a three.js scene.


Prepping the device

For this first attempt, I flashed yaw-pitch-roll sketch that's included in the imuduino github.

This sketch outputs, over btle, a string of the format: yaw|pitch|roll. Since this is guaranteed to be below the 20 byte limit for data packets, this was easier than sending the raw data and doing the IMU calculations in javascript.

In the future, I'm going to write my own sketch that outputs a compact packet of the raw data so I don't have to do string parsing, but this works for now


Client Side

The client side is pretty easy. The javascript is based on the three.js example with an added light, and a lambert material assigned to the cube rather than basic, so that the rotation is more visible.

The full source for index.js can be viewed on github. Of course, this is run through browserify because I like node's module system better than anything else, so yeah, plus browserify is awesome.

The important bits are this:

function render() {
  requestAnimationFrame( render )
  cube.rotation.x = roll
  cube.rotation.y = yaw
  cube.rotation.z = pitch
  renderer.render( scene, camera)

var socket = io.connect()
socket.on('position', function (data) {
  pitch = data.pitch
  yaw = data.yaw
  roll = data.roll

If you are familiar with three.js, you'll notice the render() function is what is needed to render the scene. In this case, we set the rotation of the x, y, z axis based on the global variables roll, yaw, and pitch.

Then, in the socket.on('position') callback, we read the pitch, yaw, and roll from the position message that the server will send.


Server Side

The server side is a very simple express + websocket server. I'm going to skip over server.js because it's messy, and it mostly just glues things together.

The one thing that server.js does that isn't boilerplate is that it converts the degrees that yaw, roll, and pitch are sent from the device to radians which three.js needs. While the client could do that, I'd rather see it done on the server.


This is where the real meat of the interaction with the IMUduino lives. The IMUDuino.js file is heavily based on nodebotanist/imuduino-ble-js by the amazing Kassandra Perch.

It's based on the [noble.js] bluetooth library, and the basic pattern that is needed is that you first discover available devices, match against a known UUID for the device, then discover the services, and then discover the characteristics of the device.

All of the necessary service ids (including my device's peripheralUUID) can be found in service-info.js

In the case of the IMUduino (and any other UART-emulating BTLE device), there is a read and a write characteristic. We only care about the read in this case.

One thing that caught me up when developing this was that the newest version of noble apparently starts up a bit quicker than previous versions, so if I tried to start scanning immediately, it was not in a powered on state. To fix this, I added the following bit of code:

    noble.on('stateChange', function (state) {
        if (state == 'poweredOn') {

This waits until the BTLE device is powered on.

The read characteristic has a read event which passes the read data into the callback function. All very node style. From here, the relevent piece is to parse the string. This isn't super robust, and I will be moving towards a binary packet design in the near future, but this works for now.

Imuduino.prototype.parse_packet = function (input) {
    var data = input.split('|')
    var p = _.zipObject(PROPS,
            _.map(data, function (n) { return parseFloat(n) }))
    this.emit('packet', p)


So we take the input, split it into an array of numbers. The next two lines are a bit of functional magic, but basically, it converts each entry of the array into a floating point number. Then, it builds an object (or dictionary) based on a second array which contains the names of the fields.

This bit is just to make the data nice to work with. I'd rather have to do position.pitch rather than position[1]because it's more readable in the first way.

After converting the data into an object, I emit a packet event, which the server.js listens for, converting the degrees to radians, and then emits a position message over websockets.



All in all, this example was a great way for me to learn how to interact with IMUduino, and how to do a little something in three.js, gluing it all together with socket.io.

Hopefully this example helps other people get up and running with IMUduino + three.js without having to run into all the same problems I ran into.

You can always reach me on twitter @RussellHay if you have any questions about this example.

"IMUduino Packs a Punch" -Atmel