Low cost LIDAR

[Maeth]

Hey all,

It is really important for us that you take a few seconds to fill out the poll! ^

I just wanted to talk about a project my team did for our final year of university.
We attended Waterloo in Canada (Woohoo!)

We wanted to make a small, low cost and easy to use LIDAR (laser based radar in our case)
We are now working on commercializing it, and selling kits while maintaining everything open source! (A la Arduino)

My question is:
Would you buy a ~$100 LIDAR that would provide you with an easy to use 2D map of a room, and also images and 9 axis IMU data?

We are also currently considering using a line laser, and hence getting 3D map data.

Check out our early prototype here:


Details are available here:
http://www.eng.uwaterloo.ca/~ibaranov
at the first link, or directly here:
http://www.eng.uwaterloo.ca/~ibaranov/files/Group_50_FYDP.zip

We welcome feedback and suggestions!

EDIT:
We will be providing examples how to use our unit, and some code, to produce full accurate SLAM such as the one below:

[Marco]

I don't play with autonomous drones so I'm not a customer ... but I'm curious, this is triangulation based I assume?

[Maeth]

Exactly, we triangulate using the known distance from the image sensor to the laser, and looking for the laser dot in the image.

[Maeth]

Some more details:

Accuracy:
We had some software issues with the prototype seen above. So we were only able to get ~0.5cm precision, 1cm accuracy over 2m radius.
However, this should be improved to 0.1cm precision, 0.4cm accuracy over 8m in production version.

Ease of use:
We will have a I2C connection, and USB connection.
The I2C will have support examples for arduino, raspberry PI, and most other popular platforms to get you going.
The I2C connection will be able to output distance and angle data, along with IMU data. This can then very easily be assembled into a map. (as above in the video)

USB will allow for more complex processing, such as depth mapping overlay on images, and creation of 3D maps with image data.
There will also be examples provided in a few languages (Python, and processing current ideas, but not really studied yet)

Lastly, the USB connection can also implement a ROS message handler, if people want one. (not sure of demand for this)

[daqq]

Hi,

The concept is nice, but there are now better alternatives. There have been some nice developments in 3D time of flight sensors such as these:

http://www.ti.com/ww/en/analog/3dtof/index.shtml?DCMP=analog_signalchain_mr&HQS=3dtof-pr
http://www.softkinetic.com/store.aspx
http://click.intel.com/intelsdk/Creative_Interactive_Gesture_Camera_Developer_Kit-P2061.aspx

The actual hardware is relatively cheap. Dunno about the software and how much is given by the sensor. And they are real time devices.

[Maeth]

Absolutely, however I think we are going after slightly different markets.

1) We do not need any processing power, all computation is done on the sensor.
This is an issue with point cloud solutions, in that you need serious horsepower to run them. this puts them out of reach of the average arduino tinkerer.

2)
They look in one direction, so you have to pan tilt them yourself

3) they do not provide IMU data or image data (image data maybe, but not same resolution (5MP) as us)

4) They are not made for mapping, more for depth sensing.

5) They still cost more (2x in some cases)

What do you mean by "Real Time"?
They still have processing time. Do you mean faster?

[daqq]

What do you mean by "Real Time"?
They still have processing time. Do you mean faster?

Real time-ish.

I meant that the aproach in the video needs to scan the area, which gives you a low update speed. The ToF sensors give you many "frames" of full 3D data per second:


I agree with them not giving you the rest of the data you provide - that's definitely a plus for you.

[Maeth]

Also, looking into it the TI solution seems to be unavailable to the general public at the moment.

They are selling dev kits, but not chips, and no actual data on chips themselves.

[Maeth]

Hey thanks for the link! 

[ChristopherN]

I would buy 3 or more for some small robotics projects. Feel free to contact me!

[Maeth]

You betcha!

We are planing to launch a Kickstarter campaign once we have a production ready design and figured out all our costs to the cent.
I want to avoid the trap of other hardware projects that go over budget.

We should be launching in September, and it is great to see people supporting this project!

[Maeth]

Any more comments people want to throw our way?

I would be glad to answer any questions. And any suggestions you make have a good chance of making it into the product, so affect the world! 

[brainwash]

I don't know much about LIDAR but my cheap ultrasonic range-finder (measurement tape) is able to get <1cm precision for anything between 0.3m and 3m, provided the surface is not overly reflective nor absorbent. I'm sure this can be improved with a more careful build. Just saying this because I think that LIDAR is overkill for these rather small distances and requires tighter computation.

[Maeth]

The issues with ultrasonic are several.

It has a wide beam, and therefore takes the average distance of several things in front of it.
For a small robot, this could be disastrous, as when it approaches a corner it will see the average of a close corner, and the wall behind it.
Lots of running into small objects that way.

Also, as you mentioned, it has issues with special surfaces. (carpet, metal, etc). Our laser will have issues with mirrors, but that's about it.

The limitations of the ultrasonic are inherent to the technology. Sound waves have to propagate in a wave front.
To improve on this would require an array of emitters to do a phased array http://en.wikipedia.org/wiki/Phased_array which would turn out to be significantly more expensive.

The last issue is range and precision, we should be able to get better values on both.

Lastly, we are currently looking into using a line laser, thereby providing a immediate 3D scan of the area. This is something you could never do with an ultrasonic.