Proposal how to proof the navigation correction by sensors

[aswin0]

I like the idea. But why shouldn't we make this into a contest. One whith more general rules. More techniques, so we can find out which one works best for localization. More freedom of models, so everybody can prove himself (m/f) as a builder.

The general rule could be: make a robot that makes 10 squares of one meter and then stops as close as possibly to its origin.

(The team (man machine) that comes closest may call themselves squareNXT champion.)

Also, no restrictions on floortype, just no markers on it.

[HaWe]

I like your idea of a contest, though it's another or somewhat different issue.
My issue is for the first step: get fundamental information about each unique method ("Proposal how to proof the navigation correction by sensors") by a standardized test model.
This is the inductive (empirical) part of investigation: collect data and check and proof thesis and antithesis.

your idea actually is an excerpt and expansion of my comparison part (3), or actually a
comparison part (4) / competition: use different sensors and combinations of sensors alternatively and additionally.
This would be the deductive part of investigation, and the synthesis.

As I made my personal investigations already a long time a go (parts 0-2, I posted the results already -admittedly roughly as it was only for my own purpose) I'm really curious about "the best method" ever.
We may discuss this in an extra thread ("competition") if you'll agree.

ps: using different floor types is sort of extra handicap or hardship; it's intended to make the robot run somewhat "out of hand" which has to be corrected by software algorithms.

[gloomyandy]

The problem with a competition is that everyones floor will be different... Still it will be interesting to see how well different people can make this work and if a few people can try it out then I'm sure the others will gain from seeing the different results... But at the end of the day I think the "real" competition here is between the different modes of operation, which means you have to be as honest as you can be. So for instance it is important to make your robot work as well as you possibly can using odometry before you start adding other sensors, that way you can see if things have really improved (it's easy to make a compass robot work better than a really bad odometry based one!).

I like the idea of using different surfaces. Another option (which I've seen in a couple of papers on this this topic), is to deliberately disturb the motion of the robot by making it run over small bumps. The example I saw used various diameters of electrical cable, and arranged things so that only one wheel would run over the bump...

[HaWe]

Another option (which I've seen in a couple of papers on this this topic), is to deliberately disturb the motion of the robot by making it run over small bumps. The example I saw used various diameters of electrical cable, and arranged things so that only one wheel would run over the bump...

a very good idea!
so how can we reproducibly define the testground ? Make a suggestion!

[loopymech]

Lego to the rescue...http://shop.lego.com/Product/?p=628

How about some Lego base plates and perhaps some other bricks or plates in the path to provide the extra bumpiness?


-Loopy

[HaWe]

I doubt that anyone will buy them just for the test.
At least I won't.
Better write your programs and start with the tests than discuss about the floor even it's just carpet or just parquett.
The floor might be changed even later.

[loopymech]

so how can we reproducibly define the testground ? Make a suggestion!

I just thought I'd respond to your last post asking for suggestions about test ground. The beauty of base plate(s) and such for building a test ground is that it can be quite clearly defined and reproduced.

But you're right...I don't think I'd buy them just for this test.

-L