Spanish boffins increase GPS accuracy by 90% Boffins at Madrid's Carlos III University have used cheap accelerometers and gyroscopes - and expensive mathematics - to improve the accuracy of GPS* by as much as 90 per cent. The team compared their results to differential GPS - which is about as good as satellite systems get - and found that with detailed analysis of the …
"Do we really need an explanation of what GPS is... in an article in The Register?"
As a brainbox on GPS, then, you can help me out with a little conundrum. If the cheapy GPS receivers in a mobile phone are only accurate to 10m, how does it (usually reliably) know which lane I'm in, given that would require circa 1-2m accuracy to give lane guidance to an acceptable degree?
Because the normal accuracy is usually around 1-2 meters in decent conditions. However 10 meter inaccuracies are not unusual and are dependent on how many satellites are in view and whereabouts in the sky they are. There are other factors too, such as ionospheric and tropospheric error and even if the satellite is in the right place and broadcasting the correct time.
It presumes you're going in the right direction on the correct road and works it out using last know speed. Easy to test, hook onto a slip road you not supposed to do, then see how long it takes to notice that you are not where you are supposed to be. You could slam on the brakes and do a u-turn and again see how accurate it is, but probably not the best on the M1.
"If the cheapy GPS receivers in a mobile phone are only accurate to 10m, how does it (usually reliably) know which lane I'm in"
Sorry, which GPS is actually giving you an indication of which *lane* you're in? It might know "northbound" vs "southbound" or similar (which is carriageway information, not lane) but it knows that purely from your actual direction. Sat navs "snap-to" roads, allowing them to be very accurate. Try coming off a road onto a farm track and watch it lose its way quite quickly.
Even when not snapping-to a known location where it expects you to be (roads), it can use fuzzy logic to determine where you are, as the 10m accuracy will vary over time, and a GPS knows that you'll roughly travel quite consistently. If your reading jumps from 5m side to side, it's more likely the error has varied rather than you've moved erratically.
Sorry, which GPS is actually giving you an indication of which *lane* you're in? It might know "northbound" vs "southbound" or similar (which is carriageway information, not lane) but it knows that purely from your actual direction. Sat navs "snap-to" roads, allowing them to be very accurate. Try coming off a road onto a farm track and watch it lose its way quite quickly.
Even better, if it tells you to take a slip road off the motorway, and you don't, it'll take quite a while to figure out. It assumes that you are following it's instructions until you exceed the "snap-to" failure threshold. Normally it takes my car until I pass the roundabout at the end of the slip road before it figures out that I've ignored it's advice because near to home I know the roads better than it does (most sat-navs don't figure traffic lights into time calculations, so while the way it sends me is shorter, it takes longer because of the stop-start traffic. So I go the long way because there are no traffic lights. Usually get home sooner)
Of course, if you wanted to get real accuracy you'd use the encrypted GPS signal only (officially) available to the military. I forget what the accuracy is, but it's sub 1 metre from memory.
I'd just be happy if the damn things gave me an indication of what direction they think I should take on the roundabout at the top of a slip road before I get to the ****ing roundabout itself.
I've made more navigational mistakes following sat navs than not. They're very useful but I also like to know where I'm going without them.
> which GPS is actually giving you an indication of which *lane* you're in?
Not quite a *lane* as such...
I use a bike computer[1] to track my flights. It doesn't snap to anything, because it's expecting you to be off-road (which is why it's useful).
It shows the track of my landings in a remarkably similar place to where I did them. Now the runway is wider[2] than a motorway lane, for sure, but the accuracy still appears to be somewhere near enough to discriminate between lanes...
Vic.
[1] It's an Android app called "Move!". I've found it invaluable for keeping a record of where I went and what I did.
[2] I can't find a reference to how wide it is - I'll ask on Sunday when I'm next there.
I think it is rather important to differentiate between a satnav device vs a GPS. I teach a ground navigation class once a year a semester at the local uni and I have to explain that while a satnav (generally) uses the GPS system to give directions, a standalone GPS only gives you your current location (unless data has already been put ino the system) and records your current location/path.
It is an important distinction. If I gave most people my Trimble GPS they would have a hard time getting from point A to B.
It looks like they are using a Novatel OEMV which in the configurations I'm familiar with will run you about $6000. An IMU on the from crossbow similar to the one they describe is another $6000. I guess it is better than the $100K systems that some reaserchers strap on their systems but not cheap.
It's inertial navigation. Nothing new really.
I've actually been wondering when car GPSes would start using inertial corrections to improve accuracy. It makes installation a little more complicated, since you have to put the accelerometers and gyros in the right place (likely slightly different for each car), but it would be a great factory feature.
The one built into my car already does it.
It knows how fast its going (speed sensor) and which direction it's going in (steering angle sensor)[1], both of which are standard on most cars these days. Given a recent fix to start from, what else do you need?
It's always amusing to watch it trying to match the maneuvers I'm making in an underground car park to the roads above....
[1] Ok, most of the time. Presumably it wouldn't have a clue what was going on if I were to four-wheel drift it round a corner.....
"The one built into my car already does it....It's always amusing to watch it trying to match the maneuvers I'm making in an underground car park to the roads above....
Yeah, the one in my TomTom does it too, but only using a built-in accelerometer.
However sometimes I have to drive into Madrid city, where they have quite a long section of motorway underground, complete with junctions. The TomTom guides you into it, and then half-way through it suddenly decides that you are on the surface, and starts trying to snap your location to the roads up there. Meanwhile, it forgets to tell you which junction you were supposed to take to get out of the tunnels in the right place.
For me issues like that, and closed/new junctions on the 2nd busiest motorway in Madrid not being updated for >2 years is more of an issue than the inherant location accuracy.
The navigation system offered in the mid 90's by Fujitsu Ten under the Eclipse brand used accelerometers and gyros that had to be installed with the rest of the system (display, "magic box", cellular antenna and CD player where you put the map disc you wanted to use). It was complicated as hell, didn't work worth a shit, very expensive and sounded like a "speak and spell" when it spoke.
I don't think aftermarket sensors are ever going to be truly road worthy. The real advances will come when most/all OEMs offer more accurate systems from the factory.
I can't wait for an autonomous car which can drive me home after I've had a few beers. We could also change from being a two-car family to being a one-car family: after the car takes me to work in the morning, it can drive itself back home then take the missus and spawn to their destination.
"outperforming differential GPS by a significant margin"
Real-time DGPS will give you centimetre accuracy, how does this system outperform DGPS?
"drift caused by atmospheric changes"
Discrepancies due to time sync differences between satellite and receiver clocks, arguably, is equally important, and definitely worth mentioning if you start explaining GPS (also uncertainty in ephemeris, satellite geometry and, being pedantic, general relativity)
Don't forget to mention that GPS is an American military system and that they can and do program the satellites to deliberately degrade system accuracy. There is also two parts to it - the open code signal that is often quoted as having 10 metres accuracy, and the secret code signal that is more accurate but not generally available to all users.
I think the Register hack was told to write a description of Differential-GPS in the footnote, but he bottled it.
1) The Amerixcan military moved away from degrading the unencrypted signal on GPS years ago. The unencrypted accuracy is the same as the encrypted. The limit is pretty much physics now.
2) They have actually removed the degradation feature from new satellites added to the constellation - as the older satellites go out of use, this option becomes less and less possible.
3) The reason the Q code is encrypted is not secrecy. It was assumed when GPS was designed that the Russian equivalent would be up and running rapidly. The Q code is there to make it harder to jam (lock on the signal when jamming is actually louder than the signal - think listening for a known tune) and to make it very, very, very hard to spoof (you would have to know the code to broadcast a fake signal).
I disagree, sir. The secret code version is about ten times more accurate, or at least has the potential to be. They used to boast about being able to put a cruise missile through a window. Also, its not difficult to jam at all.
You are probably right about them not bothering with selective availability any more.
But WTF I get three downvotes for an ordnary post like that?
I recall driving through the Limehouse Link tunnel and then between some tall buildings along London Wall one day, in 2010 I think, and noticing the loss of satellite signal, and remembering that my then-new iPod Touch could tell which way up I was holding it because of digital accelerometers. I recall mentioning this via email to a TomTom engineer ... "with a couple of orthogonally-arranged accelerometers, you could achieve inertial reference navigation, as a secondary position source when GPS reception keeps dropping out", and the reply I got indicated that this was a newly-introduced feature on top-spec models.
It's been a part of the Prius sat-nav since, ooh, about 2003. Inclinometers and gyroscopes which form part of the anti-skid system, augmented further by the high precision axle rotation sensors which the hybrid drive requires (you have to manually calibrate the nav system every now and again to allow for tyre wear which is best done on a long, straight motorway), and a position sensor on the steering shaft, which is part of the electrically-power-assisted steering.
Still that's all expensive stuff and even then brings the accuracy down to only a metre at best.
I saw a presentation by the guy from BAE who wrote the system which uses "Signals of Opportunity" (TV/radio transmitters and mobile masts), that is seriously impressive, seriously cheap and self calibrating. This proposal seems to tackle the same problem from a physical environment perspective.
In respect of the 2m accuracy killing people, I think if autonomous car designers used street maps alone for their navigation it would be very optimistic. Better to use a range of tools, like this one, combined with collision avoidance, road marking recognition, environmental awareness, etc.
"Better to use a range of tools, like this one, combined with collision avoidance, road marking recognition, environmental awareness, etc."
Yes
Might it be better to look at not hitting stuff within 2m of the vehicle as "collision avoidance" rather than navigation?
Different response times, different range.
Remember my boss telling me this: A sensor on one of the gyros failed. Decouple it, stick it in a straw-lined box (still spinning) and courier it back to the manufacturer. (Gawd knows how the driver took a corner!)
They'd get it 2 days later - stil spinning - , change/fix the sensor, spin it back up to full speed again, then return it to the boat. 2 days later, still spinning, fit it back. Bloody thing was still accurate to about a metre....Dunno if he was casting me a line, but...
I can definitely see two meters resulting in cars pushing into crosswalk areas and pushing over pedestrians, or nosing too far out into an intersection before making a turn and getting the front of the car taken off by a semi...
Stop sign icon works on a few different levels in this case......
I worked for one of the companies doing high accuracy GPS for 12 years.
With RTK GPS, you can get 20mm - yes folks - less than an inch accuracy and we could control the steering of tractors etc down to better than 150mm (6 inches). With DGPS sub-metre has been possible for many years too.
These are absolute accuracy, but when you're controlling a vehicle, drift is far more important and you re-reference from other cues.
Modern self driving vehicles would never use just GPS. Instead GPS is just one of the sensors being "fused" into an effective solution.
This is definitely not new. I know of at least one company who did work on this over 10 years ago. There are also many articles about using Inertial (gyro + accelerometer) aided tracking with GPS. One company that I know of is Cubic Defense Applications Inc. Look up "GPS denied tracking" on Google. The inertial devices being used were not high cost/expensive. They were MEMS devices...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
