Laser scanner

[Gav]

Cross post of a link from the Robots and Dinosaurs list: 

http://www.dfrobot.com/index.php?route=product/product&product_id=1125#.U2rNU62SxdX

Just in case any ROS'ers didn't see it. My writeup below; 

---------- Forwarded message ----------
From: Gav <the.mechat...@gmail.com>
Date: 8 May 2014 10:12
Subject: Re: [RnD] 360 Degree Laser Scanner Development Kit (RPLIDAR)
To: "sydney-h...@googlegroups.com" <sydney-h...@googlegroups.com>

Dang it, now I have. 

Looks like a pretty nice sensor. They say they have ROS drivers they'll release for it, which means a lot of work is saved. 

Here's a video of them using their sensor and running SLAM to build a map on the output: 

http://www.youtube.com/watch?v=uZDeNKbcnuE

It's sped up, but the quality of that map looks better than some of the 'professional' robots I've seen. (This is partly due to the laser sensor and partly due to the robot's encoders, so the laser can't take all the credit.) 

<technical mode on> 

One side effect of using common naive Bayesian reasoning like simple Kalman filters is that they assume your errors are Gaussian (bell-curves) and the readings centre around the true value. If your sensor is consistently under/overestimating the distance, you end up with weird results and artefacts on the map. 

For example lets say your sensor readings only differ by a centimeter or two, but it consistently underestimates distances by 0.1m. Let's also say your robot is 1m from a pillar. It observes the pillar, and adds it to the map (call it 'pillar-A') , but marks the location with a large amount of uncertainty, since it's only been measured once or twice. 

If the robot stays still for a bit and makes 100 measurements, since it thinks that each measurement is totally independent of the others, it marks the location of pillar-A as 0.9m away, with almost no uncertainty. After all, it's observed it 100 times, and each time it matched almost exactly what the map said, right? 

This doesn't sound like a problem, until the robot moves a couple of meters and scans again. It then sees the pillar, and checks its map to try and match it to pillar-A. Nope, too far away. It knows there's an existing pillar in the map, but it's marked that location with enough certainty that it couldn't be the one that it's now observing, since it's too far away. This is known as a 'landmark association' problem. 

There are ways around that, like not modifying the map with new measurements until you've moved a certain distance, (i.e. your measurements actually start giving you new data) but most implementations are a bit simplistic. 

</technical mode>

The sensor looks pretty darn cool. The map they made also looks clean and robust, and I'm looking forward to playing with it when it arrives. 

On 8 May 2014 08:47, Max Nippard <mnip...@gmail.com> wrote:

Have you bought one yet Gav?

http://www.dfrobot.com/index.php?route=product/product&product_id=1125#.U2q27ss_6Ai 

-- 
You received this message because you are subscribed to the Google Groups "Robots & Dinosaurs" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sydney-hackspa...@googlegroups.com.
To post to this group, send email to sydney-h...@googlegroups.com.
Visit this group at http://groups.google.com/group/sydney-hackspace.
For more options, visit https://groups.google.com/d/optout.