Is autoroute accurate?

[Panoramix]

I know I am opening a can of worms, please be considerate to each other and keep the debate civilised...

I don't own this software and I am only starting this thread as there is a lot of acrimony around distances. From reading posts and looking at it in a factual way, I think that the software may simply be wrong:

• DIY users say autoroute tend to underestimate distances
• DIY organisers say don't use Google, it will give you an optimistic distance
• Mesh users say, I don't want to switch to DIYs, it spoils my perfectly valid route which anyway is not underdistance

I just have the feeling that the authors of the software may have done some approximations which are causing us issues. We tend to assume that published stuff is right but I don't think we should.

Has anybody tried to check it on say a 10km run? A gps will give you a pretty accurate figure assuming you don't loose signal.

[Billy Weir]

*shrug*

My reading is that it is not the accuracy or otherwise of any particular piece of software that is at issue in recent discussions.  But how it is being applied.

[mattc]

Based on past discussions (and a hunch), it's not about accuracy - it's about what it chooses as the "shortest" route.
(Then there are the difference in what part of a "control town" measurements are taken from - that can make many km difference.)

It's a non-trivial problem for mapping software to calculate shortest route - much easier to measure a specific route (which is not what it's used for in this context!)

[border-rider]

My reading is that it is not the accuracy or otherwise of any particular piece of software that is at issue in recent discussions.  But how it is being applied.

My view too. What you have to do is present an org with a route they can sign off without stress.

I've learned (by trying it, by mistake) that if your route is sufficiently close to minimum distance that you're in a zone where different software can suggest pass/fail, you're cutting it too fine. 

Yes there are differences between software and the mesh distances, and they're both different from what you'll ride on the day anyway.

[frankly frankie]

We tend to assume that published stuff is right but I don't think we should.

I do, because the important thing is to establish a standard and stick to it.

Unfortunately Autoroute doesn't lend itself to this - each successive version is different, and different users probably have different preferences set up.  In retrospect it looks like it was a bad choice - but it was pretty much the only choice at the time.

This is partly why AndyC is advocating a return to - or at least not throwing out - the existing catalogue of Mesh information.  And I agree with him, even if it means reviewing it all.

Road atlases have displayed place-to-place distances in both tabular and map form, since long before I was born - and really I don't see any reason to try to improve on that model.  It's easy to bring it into the computer age too.

[Ian H]

In a limited number of experiments I set Google maps and Autoroute to measure exactly the same route. That is, using the same roads. A bit of a fiddle because it means dragging the route around and adding points to overcome the idiosyncrasies of both. The result was that actual distance measurement is about the same for both.

Then it comes down to how the software chooses 'shortest route'. I always zoom out to look at the overall map and make a visual check. If anything looks suspicious I'll move the route around until I'm satisfied it really is shortest. In my opinion the visual check is important.

The only other thing to check is the occasional odd placing of the place name points used for measuring to and from.

[Greenbank]

Has anybody tried to check it on say a 10km run? A gps will give you a pretty accurate figure assuming you don't loose signal.

Not really. A GPS is applying a smoothing algorithm to data filled with errors.

Each location point is a position plus an associated error and the GPS is guessing the path taken between them:-



All 3 of those paths (with considerably different lengths) go through the red splotches.

Looking at the tracklogs for laps of Richmond Park (just over 10km per lap) I've got more than 50 laps and with distances of 10.75 to 10.81km. So 60m variance (or ~0.5%).

[iakobski]

Those points are all good ones, but has anyone done an actual comparison between the calculated distance and the distance on the ground? It's easy to write off an overdistance as down to choosing nicer roads on the day, but it would be good to do one of:

• Ride the actual roads that autoroute specified in its measurement; or
• put in sufficient waypoints to give the actual route ridden, not the shortest route

I reckon it would have to be at least 100km to prove the point either way, rather than a short blast down a single road.

[frankly frankie]

If those 3 diagrams are all supposed to illustrate 'guesses' then the middle one looks remarkably perverse.  Not saying it doesn't happen, mind.

[Greenbank]

If those 3 diagrams are all supposed to illustrate 'guesses' then the middle one looks remarkably perverse.  Not saying it doesn't happen, mind.

No, the black lines were supposed to illustrate possible paths taken, with the red blotches the information the GPS has to calculate distance.

[iakobski]

Just to expand on what I wrote above - a long time ago I rode a DIY which came out 30% over on the ground* compared to my ancient version of autoroute**. (TBF, the DIY org did hint it might still qualify against their autoroute version with a control taken out.)

I thought this was quite a lot as the lanes didn't seem that far from the direct route, so I traced the actual route in the software and it was still showing substantially less than the ridden distance. How much better are the modern versions?


* I didn't mind, it was a nice route...
** this version included a ferry in Wales that google suggests hadn't run since the early 20th century  

[frankly frankie]

it would be good to do one of:

• Ride the actual roads that autoroute specified in its measurement; or
• put in sufficient waypoints to give the actual route ridden, not the shortest route

I'd find it very hard indeed to believe that a map-calculated distance is significantly inaccurate - along the roads it is using.
And I also think that you could check this at least as well by comparing it with documented (road atlas) distances, as you could by riding or driving it.

The issue is with what roads are being chosen, and with what points are being used as the controls.

Should we have standardised, documented single positions for all control towns and locations (I would say yes) or is it always going to be a measure ATM-to-ATM?

[Greenbank]

Those points are all good ones, but has anyone done an actual comparison between the calculated distance and the distance on the ground? It's easy to write off an overdistance as down to choosing nicer roads on the day, but it would be good to do one of:

• Ride the actual roads that autoroute specified in its measurement; or
• put in sufficient waypoints to give the actual route ridden, not the shortest route

I reckon it would have to be at least 100km to prove the point either way, rather than a short blast down a single road.

Just checked a DIY 100 up to Cambridge using Autoroute 2001. GPS of actual route taken gives 108.12km. Autoroute gives 108.3km.

[Panoramix]

Looking at the tracklogs for laps of Richmond Park (just over 10km per lap) I've got more than 50 laps and with distances of 10.75 to 10.81km. So 60m variance (or ~0.5%).

You are right, but 0.5% still is pretty accurate for the purpose of audaxing... If we were to go down to this level of accuracy we would have to start to be worried by people taking the right line and whether there are more left bends than right bends, pretty pointless IMO.

In a limited number of experiments I set Google maps and Autoroute to measure exactly the same route. That is, using the same roads. A bit of a fiddle because it means dragging the route around and adding points to overcome the idiosyncrasies of both. The result was that actual distance measurement is about the same for both.

Then it comes down to how the software chooses 'shortest route'. I always zoom out to look at the overall map and make a visual check. If anything looks suspicious I'll move the route around until I'm satisfied it really is shortest. In my opinion the visual check is important.

The only other thing to check is the occasional odd placing of the place name points used for measuring to and from.

Just checked a DIY 100 up to Cambridge using Autoroute 2001. GPS of actual route taken gives 108.12km. Autoroute gives 108.3km.

This answers my question in fact... So I was "wondering" with no basis!

We tend to assume that published stuff is right but I don't think we should.

I do, because the important thing is to establish a standard and stick to it.

Unfortunately Autoroute doesn't lend itself to this - each successive version is different, and different users probably have different preferences set up.  In retrospect it looks like it was a bad choice - but it was pretty much the only choice at the time.

This is partly why AndyC is advocating a return to - or at least not throwing out - the existing catalogue of Mesh information.  And I agree with him, even if it means reviewing it all.

Road atlases have displayed place-to-place distances in both tabular and map form, since long before I was born - and really I don't see any reason to try to improve on that model.  It's easy to bring it into the computer age too.

How can it be inconsistent and right or even a standard?

The place to place distance approach certainly addresses the issue.

[saturn]

I have a DIY by GPS planned but not ridden yet so I don't have a GPS track but....

AR2010 calculates shortest route between controls as 200.3km
If I plot the AR2010 calculated route in bikehike with follow road on it measures the route as 200.53km

AR2003 calculated a few km more, not sure whether it suggested the same route, but I had to add another control to get AR2010 over 200km (I think it gave 197.x for the original controls for which AR2003 gave 200+)

[frankly frankie]

The place to place distance approach certainly addresses the issue.

Except for people who have reservations about precisely where a place is located.
'Cambridge'
for example.

But all this is data that is in the public domain and easily accessible - type 'Cambridge' into Gmaps and you get 52.202544, 0.131237.  
If we were to agree do do it this way, there has to be some wiggle room on actual distance ridden.  It's nearly 5km from here to Cherry Hinton which is one popular Cambridge control location.

So that wiggle room agreement isn't there - and then figures of 100m (in how far?) have been mentioned - even on a 100km that would be 0.1%, ridiculous.

[Chris S]

Is Autoroute accurate?

Dunno. I don't use it. I use Viamichelin (because it's freely available across the various computer platforms I use) in conjunction with a friendly relationship with my DIY organiser  . Between us, we thrash out any distance issues (not that there have been any).

It just works.

[frankly frankie]

Is Autoroute accurate?

Dunno. I don't use it. I use Viamichelin (because it's freely available across the various computer platforms I use) in conjunction with a friendly relationship with my DIY organiser  . Between us, we thrash out any distance issues (not that there have been any).

It just works.

Bad approach.
Sorry.

[Chris S]

Is Autoroute accurate?

Dunno. I don't use it. I use Viamichelin (because it's freely available across the various computer platforms I use) in conjunction with a friendly relationship with my DIY organiser  . Between us, we thrash out any distance issues (not that there have been any).

It just works.

Bad approach.
Sorry.

Why?

[frankly frankie]

Well, because you're dealing with volunteer(s) who's time is valuable.  It's noticeable that there has been a high turnover of DIY Orgs since DIYs were invented - most regions are on a 3rd-generation of Organiser in only a few years.  Why?  Because its labour-intensive.

[Chris S]

I see.

He hasn't complained so far!  

I think Organiser workload is a different point though. By its very nature, a DIY route will always have to be checked and verified by the organiser. This doesn't have anything to do with accuracy of software.

[Martin]

Well, because you're dealing with volunteer(s) who's time is valuable.  It's noticeable that there has been a high turnover of DIY Orgs since DIYs were invented - most regions are on a 3rd-generation of Organiser in only a few years.  Why?  Because its labour-intensive.

+1 to that; I agreed to take over DIY's as they were a very similar system to ECE's which I was already doing; I had no idea just how many I'd be doing though  

the improvements to validation that John has made have been invaluable in this respect as has the online DIY entry form. Overall I'd say the paper and electronic Brevidence takes about the same time to validate.

[eck]

I've just spent half an hour plotting an entry for DIY by GPS, giving OS grid references for controls, to transfer to Autoroute so I can check it for distance.

As one of the rural idle rich, I am in the fortunate position of having the time, and I'm certainly not complaining. But I do agree with Francis that it can be very labour intensive, especially when proposed routes come up just a wee bit short: although I think organisers technically have the right simply to reject a too-short route, I've always tried to suggest ways of bringing a short route up to distance. But I also don't think that expecting organisers always to have the time "to thrash out any distance issues" is realistic. I had only about 90 DIYs to validate last year, and my AUK DIY mailbox had 1194 emails in the season just finished, but I know some organisers had a lot more than that.

I'm ok to do another year, until hopefully all these debate about AAA points, GPS validation, Autoroute etc are resolved, but I don't know if I'll be looking to do it for much longer than that.

And it's a great displacement activity, when I really should be decorating the bathroom.

EDIT: What Zoom has just said: John's recent improvements to the recording and validation are a huge improvement on the earlier version.

[Chris S]

Sounds like this needs to be taken to a new thread, as there is clearly some view-airing required on the subject of DIY organiser workload. I had no idea it was that onerous.

Returning to the subject of the thread, I doubt there is really that much difference between Autoroute and Viamichelin, and it's not currently a requirement that I use Autoroute - merely that I submit a route that conforms to minimum distance, as agreed with my DIY Organiser.

[phil d]

OK, so as a user of DIY by GPS, how can I make it easiest for my validator to deal with my rides?  I guess there are two parts to the answer, relating to entry and submission of GPX.  

While (I think) I comply with the guidelines on the AUK website, little things keep cropping up here that suggest I could do more.  Like the difficulty mentioned by Eck of translating grid references to real control locations.  I'd never considered that the use of unique references would be anything less than ideal, but it hadn't ocurred to me that because AR doesn't use grid references that's not the case.  So what is the ideal format for control descriptions?