Hacker News Reader

Comment by irrational | Hacker News Reader

One thing I really appreciate about Google Maps is how accurate the times are. There have been numerous times where google maps says is will take x minutes to get there and I think "Nah, I'll be able to get there faster than that." Nope, I've never had google maps be off by more than a minute or so. I think it must take into account not just road conditions and traffic, but also how much I am likely to go over the speed limit.

View on HN

Very much agree, with only one caveat: unpaved roads.

Google Maps estimates are wildly off when parts of the drive are along unpaved roads in Australia. It assumes a ludicrously low speed of 30-40 km/h for such roads, when most cars are able to go 80+ km/h, depending on the road. I've beaten Google arrival estimates by more than 2 hours on some drives.

When planning drives involving unpaved sections, we usually ask Google for the estimates for the paved sections (which are generally accurate), then estimate the unpaved sections based on the distance and a guess at a reasonable speed for the road.

I've also noticed that Google is much better at estimating speeds along unpaved roads that have mobile signal coverage. It seems like it uses user-generated driving data for them to some extent, but not at all for the ones without coverage. This leads me to think that they accept real-time user data, but will not "queue" the gathered data on the app for uploading later, when mobile signal is available.

This seems like a strange decision, given that unpaved roads and lack of mobile coverage correlate by nature. I suppose it's probably a security-minded decision, to prevent malicious agents from easily uploading bad data. Or maybe a quirk of the way Waze data plays into this.

It's not just unpaved roads. I also find that "private" roads (inside gated communities) around me are often extremely far off in terms of speed limit. There are private roads around here with speed limits of 45+ MPH that Google seems to calculate as 15-25MPH. It really throws off the time estimates if you're coming from or going to one of those neighborhoods. And it's been like that for YEARS. If they were using user-generated data, I'd expect that it would eventually get better, but I've seen no evidence of it. Waze does a better job at this.

It also apparently sucks if one of those private roads you turn down happens to be a private driveway and you are a car thief.

https://www.linkedin.com/pulse/google-maps-digital-trespass-...

One (totally unverified, bu) hypothesis that I have is that they drastically underestimate speed on roads that are rarely used by cars but frequently used by farming vehicles; they're using user data to estimate the speeds, they can see that the 'usual' vehicle going there is driving 30 kph but they can't see that it's a tractor pulling a wagon of hay.

I don't think the paving has as much to do with this as the smartphone traffic. I find drives on rural paved roads with no cell reception and low population are often overestimated by 30%. Maybe I drive too fast on these roads. I certainly appreciate Google keeping other drivers off these roads, however.

Interestingly enough i was really surprised when i used Gmaps in Thailand and Cambodia with the exact same issue of bad roads. It was usually accurate. However i think i always had cell signal so your assumption likely is right.

The crazy thing to me is that it's pretty decent at estimating even in dynamic traffic conditions. Consider US101 in the Bay area at rush hour - if you leave San Jose around 6pm, you'll drive straight into traffic, but by the time you get to SF it'll be significantly lighter. A lesser algorithm would give you a time prediction based on current traffic speeds at all points along your path, which would give you a wildly pessimistic estimate when you leave and only converge when you get close to your destination. But Maps has enough data to know that at that time of day traffic is trending better, and by how much.

I just figure they are using the average of other people who took the exact same trip last week when conditions were similar.

They are using people's locations in real time for traffic. >Traffic density is gathered via crowd sourcing from smartphone users using Google Maps on a mobile application in a route. In a nutshell, Google™ is analysing the GPS-determined locations transmitted by a large number of smartphone users. By calculating the speed of users along a length of road, Google™ is able to generate a Live Traffic Google Map™.

https://www.drivingdirectionsandmaps.com/traffic-conditions-...

I think it is actually not a highly complex algo. They have enough data to simply look at people who took the exact same or almost the same trip at the same time of day. Mix that with a bit of user provided traffic data to make sure it is similar to the comparisons and viola.

It's a problem that seems really simple until you dig into it. Traffic usually changes, and that's hard to predict. In broad strokes it's a "simple problem" but that's true of so many things (from bridges to self-driving cars). The complexity is in handling all the subtleties and edge cases and tuning everything so that the results actually are accurate.

There's a lot of traffic engineering theory based on viscous fluid approximation though. They'd be idiots not to leverage that.

I find that the time estimates can vary a lot when driving long distances. I think that you get a lot more variability in travel time in the countryside though, depending on driver confidence and experience.

I usually manage to knock a 4.5 hour drive down to 4 hours or less.

Their acquisition of Waze really gave them the edge here. Waze has been getting better and better at predicting times. I find that in most situations it is rarely off by more than a minute or two.

I have the same opinion about their acquisition of Waze. But I think Waze itself as a GPS is quite frustrating. It tells me to take the worst local road, when Google Map takes me the better route (the exact route I would have taken anyway without the GPS). After tryig a few times I have given up on Waze completely. I feel Google is using Waze for getting additional usee data points.

I also found them to be crazy accurate for walking directions. I always think "well, I fancy myself a speedy native New Yorker amongst all that gawking tourist data Google must have, I can cut down this half hour walk by a few minutes at least". Almost always arrive right on their estimate.

Interesting, Ive had the opposite experience while driving back to NYC from other east coast cities. Google generally takes into account that traffic will increase as I approach NYC, but the initial estimate is always off by at least 1-2 hours.

I wish they introduced a "motorcycle" vehicle type. At least for cities, it makes a big difference in arrival time estimations.

They did that recently in India - not sure if and when they'll roll it out to the rest of the world.

I sometimes got the impression they do but still present it as car. I've got highly unrealistic driving times of a car in Bangkok that were very close to my actual driving time on a scooter for example.

Seems tougher to do estimation specifically for motorcyclists.

Some areas allow lane splitting, others don't. Some riders are on tourers that can't hop through traffic but can maintain highway speeds comfortably, others ride single-cylinder dual sport bikes that can't maintain 60 without taking a physical toll on the rider. And every rider has a slightly different risk tolerance, which translates to different behavior in traffic. I personally don't mind lane splitting between two semi trucks cruising abreast at highway speed, but I get anxious squeezing through cramped city traffic. And I slow down considerably anywhere a SMIDSY might happen, even if I have the right of way.

Also, it may be difficult for Google to identify who is riding a motorcycle from location data alone, which is what they would use to generate the motorcycle estimations. Not sure how they would do it...maybe using accelerometer/gyro data to identify inward leans during turns? The assumption being that cars will lean outwards, while bikes will lean inwards.

Reply on news.ycombinator.com