I was discussing with a friend the application of GNSS in the transportation industry. The forward direction accuracy might satisfied by a common car GPS unit, or perhaps some modest improvement would be beneficial.
?ÿThe difficult requirement is to know what lane, or which of parallel railroad tracks, it is in with near certainty. Thus, in the lateral direction, rms (sigma) or 95% confidence values are mostly irrelevant. The spec is in terms of how many nines you can put in 99.99??% confidence at a chosen distance on the order of a meter or two.
?ÿThere is some fear that, out in the tails of the distribution, the gaussian normal model won??t apply and you can??t depend on getting 99.999% at 4.42 sigma, or whatever you would calculate. In other disciplines, statistical tails have often been found to not fit the normal model.
?ÿTime averaging must play some part in this. A near-real-time result is needed, but there may be benefit for some applications in allowing a lag of a few seconds that would help fight multipath ??excursions? in the reported position.
?ÿHas anyone seen actual statistics on the extremes of GNSS position data?
?ÿWould something similar to this performance be possible with corrections from CORS or would it take closer-spaced reference stations?
MN DOT studies might have that information.?ÿ They were first published about 10 years ago, studying snowplow guidance systems.
Greater precision comes with something better than L1 with SBAS corrections. That is the purpose of L5 and the broadcast of L5 corrections from SBAS satellites as well. The transportation industry has skipped over L2.
The next step would be to launch GPS/SBAS satellites in polar orbits. If I recall correctly China has Bediou satellites in polar orbits and Japan has SBAS satellites in polar orbits.
3 sigma, 99.7% is sufficient, but it may have to be a simple mean of 2 sigma L1 and 2 sigma L5.
For railroads it is a linear position along a path that is 4.5m from an adjacent path.?ÿ For a train leaving the track it is easier to recognize the skewed path over the deviated position.
I am not familiar with any Glonass SBAS corrections, but Galileo and Bediou corrections are operating.
Paul in PA
?ÿ
3 sigma, 99.7% is sufficient
That doesn't seem good enough.?ÿ An out-of-lane indication for an average of 10 seconds every hour would seem problematic.
Even 1m may not be out of lane. That 0.3% is out of specified precision, not necessarily out of lane.
Paul in PA
?ÿThose folks developing that stuff are way smarter than I am, but here's a back-of-an-envelope run at quantifying the problem. Suppose that we have an 80-inch wide automobile traveling in a 144-inch (12 feet) wide lane. If the auto is in the center of the lane, there are 32 inches of lane variance on each side. For a 3-sigma solution, sigma would have to be 32/3 = 10.67 inches and for a 4-sigma solution, sigma would be 8.00 inches, and so on. Of course, not all lanes are 12 feet wide and semis are 96 inches wide (sometimes 102), so an envelope back won't suffice for the analysis.
According to this article, Mitsubishi is launching its own satellites in geosynchronous orbits?ÿ https://www.japantimes.co.jp/news/2016/03/30/business/tech/missile-maker-adapts-guidance-systems-self-driving-cars/#.WjGV4_mnEzw
And according to this one, they're getting very serious about the prospect?ÿ http://www.mitsubishielectric.com/news/2017/1027.pdf
Gives new meaning to "Leave the driving to us."?ÿ
?ÿ
sigma would be
But are the statistics of GNSS accuracy accurately modeled out to several sigma by the gaussian-normal curve?
?ÿ
I don't know, but it's a relevant question. I can't get past what the instantaneous analysis to decide if a measurement is good or not would be.
I would think you would link several systems. Radar for accident avoidance. Some kind of stripe detection system for lateral movement I know Chevy has a system that beeps if you cross a line. I would think GNSS would be used more for speed and general positioning than the fine steering.?ÿ