indoor / outdoor positioning system?

[gloomyandy]

You may have sen this already...
http://www.youtube.com/user/gloomyandy? ... MI63k5W3sU
It uses a modulated laser and works pretty well in bright light. But it would probably only work over a range of 8m or so. I suspect you will struggle to find anything that will meet your requirenments of location size and nxt ready. The outdoor and large area requirements are pretty tricky combinations, but it would certainly be interesting if a solution was available...

[HaWe]

What update speed do you need, and what update speed do you want?

depending on the following accuracy within a 10m detection range:
either: angle triangulation: 0.5 degrees for 3-4 beacons, 1 degree for 6 beacons
or: distance detection: 10cm / 10m = 1 percent

- if I got a quick tracking system (odometry) maybe once every 10-20 seconds
- if I have no odometry (as currently) and no stochastic filter: 10 times every second
- no odometry but a particle filter together with gyro and compass (and maybe an accelerometer): maybe once every second

(just a guess)

@andy:
your video is pretty cool! - higher detection speed (while moving => less accuracy), modulated laser beam (higher specifity), and combined with a particle filter to reduce noise:
could be perfect.
Even with RCX encoders instead of the built-in NXT encoders...

[linusa]

Doc, we talked about this a while back ago, but you deleted my post with all my ideas (and unfortunately I don't have a copy to post). Good thing you can't do that here .

Well, currently I'm using this system: http://naturalpoint.com/optitrack/ , and I'm very happy with it. (To be precise: these cams). The also sell much cheaper systems, see http://www.naturalpoint.com/trackir/ . If you need advice on IR interference or precision or anything else, ask.

The bottom line is: Either buy something,
or
do something yourself.

Optical tracking is probably best. But you need software or course. If you don't want IR, you can use normal webcams and colored patches -- pretty ok, if you make sure the color markers don't appear anywhere else. Or you should use something like this: http://www.hitl.washington.edu/artoolkit/ . Just one webcam needed, pretty good results.

Another alternative: Kinect! Get one, 130 Euro, well spent money. Unfortunately, you need to write software again with certain complexity, and max range is about 5m and there might be certain outdoor problems.

All of those suggestions can reach the < 5cm precision you need.

[linusa]

If you just use a simple inexpensive webcam (< 40 euros), you can get already amazing results. You can also manually remove the IR filter and then have a near-IR cam. One impressive example: http://www.youtube.com/watch?v=OD6i_fkT1ZE (the amazing thing is, you can read the guy's forum posts in German somewhere, and you can see how he developed this step by step without prior knowledge in the field of computer vision).

[linusa]

And one last post on that topic. You can also use two (or more) Nintendo Wii remotes, I already posted about this here. They cost about 30 to 40 euros each, less when bought on ebay. They work via Bluetooth, but in realtime and really cool (unlike the NXT). The can track up to 4 IR sources with 100 FPS. See this website and the video http://www.cl.cam.ac.uk/~sjeh3/wii/ . I tried that system and re-coded it, it works with cm-precision, range up to 4-5m (maybe more with more powerfull LEDs).

You can also mount the Wii-Motes on your robot and have the markers fixed somewhere (as beacons). The Wiimote is battery powered and its Bluetooth range should be even greater than the NXTs. An example of how that could work is here: http://www.ra.cs.uni-tuebingen.de/forsc ... ome_e.html

[HaWe]

I can only recall some posts of yours where you replied completely off-topically not to the TO question.
which of the mentioned systems of your upper postings (inclusively bearing or optical tracking) can you directly plug to the brick and run it's navigation software directly on the brick ?
(It has to run on an autonomous NXT without any PC or attached additional microprocessors. I specified that in a former thread, I missed to do this in this thread.)

And how does the positioning work indoors and outdoors in a 150 m² area with which accuracy? (0.5-1 degrees or percent of distance up to 10m is needed as I wrote)

[mattallen37]

~Helmut, there is no perfect solution. You are gonna need to settle for less than perfect. Linus was just posting what he thought could have been inspirational for you.

[linusa]

I can only recall some posts of yours where you replied completely off-topically not to the TO question.

Discussing the combination of different systems and ways to identify beacons (modulation, pulsing, etc), as well as triangulation, isn't off topic in my eyes. But anyway:

which of the mentioned systems of your upper postings (inclusively bearing or optical tracking) can you directly plug to the brick and it's software run directly on the brick ?

What you mean by "directly"? Depends on the definition! If you buy a gyro, you can "directly" plug it into your NXT, but you have to write software to initialize and use the sensor (in this case, several lines of NXC code -- not much).

If you have a computer or netbook and send data via Bluetooth, it can be "directly" received by the NXT brick.

You can directly wire the Wiimote's IR camera to the NXT brick, as it speaks I2C. But the software and protocol might be challenging (as I recall that certain I2C commands on the NXT are difficult to implement).

I've seen mobile robots having a Kinect on board -- I've even seen a flying quadrocopter with a Kinect. But then you need USB wiring and chips and hardware and drivers.

Apart from that, image processing on the NXT is a thing I'd rather not do. The NXT doesn't have enough CPU power. You need the data at least preprocessed (like Kinect's or Wiimote's output, or the Optitrack cams).

And how does the positioning work indoors and outdoors in a 150 m² area with which accuracy? (0.5-1 percent at 10m distance is needed as I wrote)

What volume are you tracking? 150m² and what height? If your robot is confined to the ground, it gets a lot easier. I've got a copter in about 3m x 3m x 2m or maybe 4m x 4m ground space. The Optitrack cams are ~ 1mm, but depends on number of cams (and other params).

I've seen ARToolkit beeing used in about 3 to 4m distance with good (< 5cm) accuracy. You can scale this, BUT your code-sign needs to be bigger. Is that ok? Having a 1m x 1m paper sign on top of your robot? How high can you put your cams? Do you get a bird's eye view?

A good approach for this large base area of 150m² can be hand-over: Have several 25m² areas with separate tracking. Then scale it up! Advantage: You can adjust camera settings to bright parts of your lawn (direct sunlight) and darker parts.

Oh, and there will be no complete tracking solution for "indoors + outdoors + large area + precise + NXT compatible".

You want simple answers, but there are none. You don't want to go somewhere and say "I'd like to have one complete tracking solution, please", or do you? This is engineering, not shopping. So there will be problems we all can't think of right now. And professional tracking solutions which can cope with varying light conditions and all that cost > 20k euros, easily up to 100k.

[linusa]

(It has to run on an autonomous NXT without any PC or attached additional microprocessors. I specified that in a former thread, I missed to do this in this thread.)

The NXTcam has additional microprocessors included, does that count? The Wiimote camera can be talked to via I2C, but internally it has image processing and a Kalman filter on board. Does that count? You can make things arbitrarily difficult by prohibiting things. Ok, this excludes webcams and any other cameras.

We slowly come back to the things I proposed a while back ago, and which are similar to the Laser triangulation video. Pulsating laser pointers + IR seeker sensors (maybe turning) are the way to go. You could use 2 laser pointers with 2 polarization filters, and 2 IR seekers with polarization filters as well. But no, I can't prove to you that it works. But I made good experiences with horizontal and vertical polarization filters. You loose 50% the intensity, but it really works well as in: You only see the correct source.

[HaWe]

no, built-in µCs are ok if the sensors are premanufactured like the MS cam and could be plugged like that directly to the brick without having to use additional bradboards or so.
I can use a Wii-mote if it could be plugged directly to the brick without soldering, if not then: not. Also for pulsating Laser there must exist a premanufactured Lego-compatible connection interface. No soldering, no bradboards.

But also additional NXT bricks are allowed (plugged as slaves to the BT master or via RS485).
But no external intermediate Atmels or PICs - or PCs.

I can not write i2c drivers for any hardware, but I can write navigation software using an existing driver-interface (like HS IR seekers or a compass or a gyro ar an accelerometer for example). Also for a cam if it was well-documented with software examples (not as poor as the MS cam).

I don't need replies like "the answer is : there is not such thing and not other on the NXT, take a PC with MATLAB"... (I slowly recall why I once deleted some posts in the German mindstorms forum some years ago) :?

But the good news is:
It has NOT to be cheap! ^^