SLAM for hobbyist robots

Posted by John Nagle on February 13, 2010, 1:08 pm

Hobbyist robots are still too dumb. We're still seeing battery-motor-wheels,
maybe with a few sonars. I had one of those 25 years ago. CPUs have advanced
only from the Basic Stamp to the Atmel AVR, often in Arduno cult form.
This is too lame for 2010.

A logical next step is to get visual Simultaneous Location and Mapping
("SLAM") onto hobbyist robots. Then they know where they are and what
the environment is like. A cell phone with a camera has enough CPU
power for this. The published papers are out there. Willow Robotics
gives out some SLAM code.

Anybody doing this?

John Nagle

Posted by D Herring on February 13, 2010, 2:07 pm

On 02/13/2010 01:08 PM, John Nagle wrote:

> Hobbyist robots are still too dumb. We're still seeing
> battery-motor-wheels,
> maybe with a few sonars. I had one of those 25 years ago. CPUs have
> advanced only from the Basic Stamp to the Atmel AVR, often in Arduno
> cult form.
> This is too lame for 2010.
> A logical next step is to get visual Simultaneous Location and Mapping
> ("SLAM") onto hobbyist robots. Then they know where they are and what
> the environment is like. A cell phone with a camera has enough CPU
> power for this. The published papers are out there. Willow Robotics
> gives out some SLAM code.
> Anybody doing this?

As I understand it, the SLAM systems in many commercial mobile robots
aren't overly robust. Many rely on a 2D or 3D sick laser or the like.
So its not surprising this hasn't infiltrated hobby systems.

The best resource I know of is
http://openslam.org/
but its been a year or two since I last looked very hard.

- Daniel

Posted by waynegramlich on February 13, 2010, 3:04 pm

John Nagle wrote:

> Hobbyist robots are still too dumb. We're still seeing
> battery-motor-wheels,
> maybe with a few sonars. I had one of those 25 years ago. CPUs have
> advanced only from the Basic Stamp to the Atmel AVR, often in Arduno
> cult form.
> This is too lame for 2010.
>
> A logical next step is to get visual Simultaneous Location and
> Mapping ("SLAM") onto hobbyist robots. Then they know where they are
> and what
> the environment is like. A cell phone with a camera has enough CPU
> power for this. The published papers are out there. Willow Robotics
> gives out some SLAM code.
>
> Anybody doing this?
>
> John Nagle

At the last HBRC (Home Brew Robotics Club meeting, somebody
(alas I'm not good at names), showed up with a robot that
was going to do camera vision based SLAM.

One of our members showed up with a pre production version of
the Neato floor cleaning robot, which uses a laser range
finder (not Lidar) to perform SLAM. The environment at CMU-SV
(Carnegie Mellon Silicon Valley) was pretty hostile, so it
lots of problems with chairs everybody was sitting in.
The chairs have 4 thin chrome plated steel legs, so they
behave like small cylindrical mirrors at visual wave lengths.
The Neato does have a bump sensor, so it did not totally fail.

What is enabling all of this is the deployment of low
cost and low power 32-bit platforms that can be placed
on the robot. OpenCV (originally from Intel, not maintained
by Willow Garage) is getting a lot of traction these days.

I keep hoping that one of the low cost 32-bit platforms
out there will get some traction like the Arduino did.
32-bit CPU, Linux, Wi-Fi, some USB ports, and some lower
level stuff like Serial, I2C, SPI, some A/DC, should be
less than $50 these days.

-Wayne

Posted by John Nagle on February 14, 2010, 1:56 pm

waynegramlich wrote:

> John Nagle wrote:

> What is enabling all of this is the deployment of low
> cost and low power 32-bit platforms that can be placed
> on the robot. OpenCV (originally from Intel, not maintained
> by Willow Garage) is getting a lot of traction these days.

The Willow Garage people have cleaned up OpenCV. See
"http://www.willowgarage.com/pages/software/opencv" . A few years ago,
I tried OpenCV's stereo code, and it was awful. For example, about 10% of the
time, the camera calibration program would produce total garbage. The
stereo algorithm never worked very well. The LK tracker was good, though.

Willow Garage has been doing a lot of work on "astereomatching.cpp".
It's worth looking at that again.

John Nagle

Posted by waynegramlich on February 14, 2010, 7:48 pm

John Nagle wrote:

> waynegramlich wrote:

>> John Nagle wrote:

>> What is enabling all of this is the deployment of low
>> cost and low power 32-bit platforms that can be placed
>> on the robot. OpenCV (originally from Intel, not maintained
>> by Willow Garage) is getting a lot of traction these days.

I meant to say "now maintained" rather than "not maintained".

> The Willow Garage people have cleaned up OpenCV. See
> "http://www.willowgarage.com/pages/software/opencv" . A few years ago,
> I tried OpenCV's stereo code, and it was awful. For example, about 10%
> of the
> time, the camera calibration program would produce total garbage. The
> stereo algorithm never worked very well. The LK tracker was good, though.
>
> Willow Garage has been doing a lot of work on "astereomatching.cpp".
> It's worth looking at that again.

We have two or three people in the club having some level of success
with OpenCV these days.

-Wayne

Page 1 of 2 1 2 > last >>

Subject
Posted In
Date

Linux drivers for robots: /dev or sockets ?
General Robotics Forum
2009-10-19

The trouble with hobbyist-level robot hardware
General Robotics Forum
2006-08-17

VisRecog module of MindForth AI for robots
General Robotics Forum
2011-09-23

Linux router as embedded computer for robots
General Robotics Forum
2010-01-23

SLAM for hobbyist robots - it's time.