It's called differential GPS. However the new block of GPS sats have higher accuracy without the need for the (multiple) expensive receivers.
However, I'd argue the lanekeeping and "where am I" problems this stuff solves is dwarfed by the common sense and logical reasoning & recognition problems.
GPS is only accurate to a certain amount (maybe 1.5 metres, maybe 10 metres) in an absolute sense, but in a relative sense they are much more accurate and would easily show circles being driven in a parking lot (it's just that the circle might be 10 metres away from the road).
This is also used in surverying to do something called differential GPS with two receivers: one GPS receiver fixed in a known location, and another to measure GPS locations relative to the first GPS in the known location - absolute error is completely removed by this process (http://en.wikipedia.org/wiki/Differential_GPS claims 10cm accuracy). If you think about it, two receivers separated by a little space is more or less equivalent to two receivers separated by a little time, so you'd expect a similar relative accuracy of 10cm.
The highest resolution GPS receivers have two changes:
1. Base stations at known locations broadcast corrections for factors like the effects of the ionosphere and troposphere, and inaccurate satellite ephemerides. If you have multiple base stations, you can interpolate between them (Trimble VRS Now [1] does this, for example)
2. Precise measurements by combining code phase (~300m wavelength) and carrier phase on two frequencies (~20cm wavelength), plus the beat frequency of the two frequencies (~80cm)
With these combined, and good visibility of the sky, centimetre-level accuracy is possible.
Autonomous vehicles will often combine this with inertial measurement [2] which offers a faster data rate, and works in tunnels.
Many people also expect autonomous vehicles to also track lane markings, in combination with a detailed map to say whether lane markings are accurate.
WAAS has been available on consumer GPS devices since the mid 2000's and is not "professional" nor does it address reception/interference issus; WAAS provides semi-real-time position refinement data based off drift observed by ground stations within a certain range to compensate for things like atmospheric disruption of the signal. WAAS is built-in to the vast majority of receivers these days.
Multi-band GNSS is what helps address issues around reception like multipath and interference. Receivers used to be pretty expensive, and the number of L5 satellites wasn't that high until the last few years, but now can be found in even a number of consumer electronic devices, and the number of L5 satellites keeps going up (and as more satellites have L5 signals, things will get better.) Several Garmin sport devices have it now, for example. A quick search shows Maxim sells a multi-band GNSS chipset for $10.
There is really no excuse for any autonomous vehicle company to not be at least evaluating multi-band GPS receivers, if only to give more reliable position data in urban environs.
I'd be surprised if at least some aren't using RTK receivers, but this requires a base station within a certain range of the cars, or commercial service (such as that offered to farmers by John Deere.) RTK just isn't that expensive compared to even a single LIDAR unit.
Edit: they can tolerate at least some degree of GPS degradation and use dead reckoning / lidar data in some cases, according to this person: https://news.ycombinator.com/item?id=31940167
What's even more amazing, is that differential GPS works at all, given that it relies on measuring the timing of the underlying carrier signal. (As opposed to the pseudo code.)
It takes the advantages of digital signaling, and throws them out, instead measuring the analog carrier waves directly. That this can be done at such low signal levels is unbelievably impressive.
(The statistical methods involved in getting an initial signal lock are nearly as impressive.)
And the whole thing was a 1980s hack for land surveyors to get better than 100m accuracy while systematically eliminating uncertainty due to Selective Availability (SA).
It's a hack that improved the overall accuracy of the system from ~100m to sub ~1cm, and eventually sub ~1mm accuracy, when used with some other methods.
I'm pretty sure if you're on the Earth's surface and can pick up three GPS satellites, the other solution is out in space. If you see four or more GPS satellites, there's a unique solution.
And then there's RTK (I like to talk about https://centipede.fr ) that measures the offset between GPS signals and litteral ground truth and proadcasts it, enabling centimeter-level precision.
The more systems, the merrier, even if they aren't all accurate, you can use them to increase precision (kalman anyone?).
Someone mentioned multi band. You can also use signals from multiple systems at different frequencies.
I don't think differential GPS does anything here. It corrects for large-scale ionospheric delay error, which is going to be exactly the same for every drone at this scale. And they have WAAS receivers anyway, which provides better accuracy than differential GPS. (Maybe you mean RTK, which could help achieve centimeter accuracy.)
However, I'd argue the lanekeeping and "where am I" problems this stuff solves is dwarfed by the common sense and logical reasoning & recognition problems.
reply