Kalman filters: Controlling ROVs and AUVs

You may want to try a more modern approach - H infinity state estimation and H infinity control. Only slightly different from the mse approaches they minimize the max value of the error. They are more robust to innacurate models and work on deterministic noise sorces even. In fact no knowledge of the noise is needed at all as compared to say the Kalman filter (though weighting must be selected based onSNR).

of typos on this article but you will get the jist)

The author of this paper also authored the book "Optimal State Estimation". He starts you on Kalman filtering, takes you through H-infinity filtering, then on to some modern semi-esoteric stuff that requires orders of magnitude more processing power than is available now (but who's to say it still will ten years from now?).

I know you're not a total newbie to DSP, but if you don't have much control background you may also be interested in my book (see

It only covers basic control theory, but it takes you from that basic theory all the way to practical applications using today's technology -- if you haven't closed real loops in real computer hardware I think you'll find it useful.

(see

It only covers basic

Both books are on order from amazon.com.

I've been browsing IEEExplore, but I don't have access to the relevant journals and don't know what to look for. I suppose what I need is similar to aeronautics nav processing, where one fuses data from onboard sensors, sattelites and beacons to produce a position/track estimate.

Any suggestions on where to start?

Rune

Rune, I'm not sure it's directly relevant, but there is an article from the airborne laser scanning (ALS) community by Baltsavias:

Baltsavias, E.P., 1999. Airborne laser scanning: basic relations and formulas. PRS, Vol. 54, No.2-3, pp. 199-214.

ftp://igpho.ethz.ch/pub/manos/papers/formals_prs.pdf

that ties together some things there.

ALS fuses GPS, navigation systems, the laser scan angle and time-of-flight measurements to get 3D data. It's a different problem, but I suspect there's details in the literature that will help you out.

Ciao,

Peter K.

(see

It only covers basic

I think you were on target (in general, at least) in asking about Kalman filtering for this part. The definition of Kalman filter has taken on the meaning of "a multivariable signal processing algorithm that takes data from a big wad of inputs and coughs up the best output(s)" -- in this wide sense you were asking for just the right thing, at least inasmuch as H-infinity, extended Kalman and particle filters all fit the definition as well as a plain old Kalman filter.

Dan Simon's book should prove very helpful for you here. I know enough about Kalman filters to construct one to the narrow definition, so when I have a problem that's narrowly defined enough I can be useful (and if you're conservative in your choice of weightings you can make it generally useful, although not necessarily optimal). The next time I have occasion to do a multivariable estimation problem I'll be leaning on the Simon book, and I expect to be able to do a pretty good job of it.

When (or if) you get to the point where you're actually controlling the motion of the vehicle you'll find my book useful, although I imagine you'll have to concern yourself with all sorts of nonlinearities and axis cross-coupling that one can normally avoid in simpler control problems.

Rune Allnor wrote in news:1182749246.005438.80620 @w5g2000hsg.googlegroups.com:

Hi Rune,

No satellite navigation (GPS) when submerged!

Sam

No, but if you can surface & stick an antenna up you can get a fix, and use it to calibrate your inertial sensors.

You'd be able to get LORAN down a few meters at least -- I'm not sure how far it'd penetrate the water, but if they can use 30kHz to communicate with subs 100kHz should get down some.

Given a compass, inclinometer and a water-speed gauge you could use dead reckoning, particularly if you could surface every once in a while to upgrade your knowledge of drifts.

Rune -- I think you need to post this to yet another newsgroup. Find the one that covers choosing the right sensor technologies for AUV's and you'll be fine...

subs use a floating antenna for LORAN

Well, there's nothing like practical knowledge to blow theory all to hell.

Come to think of it, a GPS receiver on a tether may work, also, if you are working that close to the surface. It adds some interesting twists to your "Kalman" filter, to be sure, but it may work.

Is that technology not used to track the movements of some marine animals, like whales?

I know.

However, the general setup is very similar aeronatics:

When using ROVs, a surface vessel (with GPS nav) serves as a "moving refernce" much the same way GPS satellites do in aeronautics.

A number of refernce transponders have been installed in the area we operate, serving as fixed beacons for both ROV and AUV.

The vehicles have onboard INS systems.

The trick is to make sense / use of all these nav data flowing around the system.

Rune

It used to be a standard excercise in the course in underwater acoustics to compute the attenuation of EM waves in salt water. If I remember correctly, EM waves are attenuated by some 50 to

60 dB per wavelength. Given a frequency and signal strength at the surface, one can compute how far down a given system can recieve a useful signal.

This is a reason why subs communicate in the ELF range.

Rune

I guess that depends on which part of the problem you really want to solve yourself. The Kearfott T-16 and T-24 units that we put on our AUVs take in the ADCP and GPS data directly, along with our acoustic fixes, and does the sensor fusion itself. The output is its best estimate of position, based on its own filter. After we manage to convince it of the correct starting point, it does very well as long as ADCP bottom lock is maintained; I think about 0.2% of range, but don't quote me on that.

I amd not designing a system, I am evaluating one.

The expected accuracy of our surveys is very high. Whenever we are out of spec we have to do a system evaluation to find out why, and if there were anything we could or should have done differently. I know a lot about the acoustcis side of both surveying, comms and oceanography, and have developed a more or less trustworthy instinct for how those sorts of factors affect the accuracy of the surveys. Now I need to learn as much as possible about the navigation side of things. I can't do the required evaluation unless I know in as much detail as possible how the C&N systems work, what their "intrinsinc" characeristics are, and how they are affected by oceanography, platform stability, only occasional access to GPS references, sparse / dense reference transponder grids, etc.

Rune

Hi Rune, SAR historically had similar problems blending GPS and INS systems and compensating for Lever Arms. Some of that problem has gone away with the newer technologies. You could check from either DRDC-Ottawa (Mostly Radar) or DRDC-Atlantic (Mostly Sonar). Here's the basic link:

You could also try our friends at FMV - The new Visby class has ROV's, so they may have looked at some of this. Perhaps contact Elias.

Cheers, David

Hi David,

I talked with Elias on the phone a couple of weeks ago. He clicked onto this immediately; I'll probably use some of my vacation to go visit him some time in September. We are coming up on the Real Deal now; 6 times 120 km of surveying to be performed inside, say six weeks or so. The "make or break" kind of thing.

Oh well. I'm a mere 18 hours short on a 4-week trip, the second completed inside of ten weeks. A shore-side vacation will be very, very nice now...

Rune