How Accurate is HTML5 Geolocation, really? Part 2: Mobile Web

Where Part 1 focused on non-GPS enabled devices, Part 2 is totally focused on mobile web geolocation. The great news is that the usage of HTML5 location services along-side the fact that there is a GPS chipset in most, if not all, modern smartphones and tablets dramatically improves the chances of getting an accurate location. And, besides that fact — mobile geolocation is simply a lot of fun to work with.

I also want to point out that there are an increasing number of really good blog posts covering the topic of “how to use” the API that look at the nitty-gritty of how the code works. This post is different in that I’ve tried to focus on “how to build successful applications” with the API, and how to get the most out of the API so that you can successfully implement your unique requirements.

What’s different about desktop vs. mobile HTML5 Geolocation? With mobile you can access the GPS if it’s available. It’s important to note that in order to access a device GPS you have to set the optional enableHighAccuracy property in your code. Contrary to what is shown in some samples on the internet, you can use this property with both the getCurrentPosition() and watchPosition() functions.

//One time snapshot
navigator.geolocation.getCurrentPosition(
     processGeolocation,
     // Optional settings below
     geolocationError,
     {
         timeout: 0,
         enableHighAccuracy: true,
         maximumAge: Infinity
     }
);

//Tracking users position
watchId = navigator.geolocation.watchPosition(
     processGeolocation,
     // Optional settings below
     geolocationError,
     {
         timeout: 0,
         enableHighAccuracy: true,
         maximumAge: Infinity
     }
);

How accurate is it??? This is the million dollar question, right? When using enableHighAccuracy() on a phone where all the appropriate permissions have been selected and granted, I’ve typically seen accuracy readings as low as 3 meters (~10 feet) that were obtained within 10 – 30 seconds of kicking off the geolocation functionality. I’d consider that excellent for most consumer and retail applications. You should be aware that like any location-based functionality you will get spurious (abnormal) results that fall way outside the norm, and sometimes these results are wildly wrong.

I’ve seen claims that using the enableHighAccuracy() property slows down the phones ability to deliver a location. I’m going to argue that those claims are misleading. It is true that the GPS, itself, can take a significant amount of time to warm up and start delivering high accuracy results. For an in-depth look at that topic see my post on the Six Most Common Use Cases for Android GPS. However, there are conditions where simply enabling the enableHighAccuracy() property doesn’t affect the speed in which you can get the initial result. More on these topics below.

What is the best way to try out various configuration scenarios? I’ve built an HTML5 Geolocation Testing tool that can be used in browser, or it can be repurposed to work in PhoneGap or Titanium. It is a jQuery-based mobile application that includes a map and settings view were you can adjust all the different properties and try out different configuration scenarios. It’s a work-in-progress so I welcome suggestions and pull requests.

 Why HTML5 Geolocation rather than native? Applications using HTML5 Geolocation typically have slightly different requirements than native GPS-based applications. Each platform has its advantages and disadvantages and it all comes down to your requirements, budget, timeframes and skill sets:

  • Ability to re-use existing JavaScript and HTML5 skills to build a high-accuracy mobile application.
  • Don’t have access to native platform developers or skillsets on Android, iPhone and/or Windows Phone.
  • Need a cross-platform stand-alone web app, or a web app that has been repurposed to work with PhoneGap or Titanium.
  • Quickly locate the user/consumer within a reasonable expectation of accuracy.
  • Typically it is a non-commercial, consumer grade application that does not have extremely high accuracy requirements (e.g. < 1 meter).

How fast can I get an initial location result? The answer is very fast, potentially within a few seconds, given the following scenarios:

  • If there was a cached GPS or Network location stored on the phone. The GPS location is, of course, from the GPS chipset. The Network location comes from your wireless carrier and is dependent on your phone and their capabilities.
  • How the timeout and maximumAge properties are set. If you set timeout = 0 and maximumAge = Infinity it will force the application to grab any cached location, if one is available. Other settings may result in delays.
  • If the phone or tablet has decent internet connectivity and Wifi enabled.
  • If the device is in an urban area with many wifi nodes broadcasting their SSIDs nearby.
  • The device has a clear and uninterrupted view of the sky. GPS’s listen for a very weak signal from multiple satellites. These signals can be partially or completely blocked by buildings, thick foliage, vehicle roofs, etc.

 How accurate is the initial location result? Hah, you might have guessed I’d say that it depends.  When you first kick off a geolocation request, accuracy does depend on a number of different factors that are mentioned above. And it’s safe to say that, in the vast majority of cases, the first location is not the most accurate and typically not the most dependable. If you want the fastest, most accurate location possible then you will most likely need to either do multiple snapshots, or use watchLocation until your desired level of accuracy is met. It’s important to note because I’ve been asked about this many times, you cannot expect the GPS, itself, to have enough time to lock onto a satellite and deliver a fast, accurate initial location. It may take dozens of seconds or even minutes. Yep, it’s true. Factors that affect initial location accuracy include:

  • Cached locations – how recently the user accessed location functionality. For example, applications like Facebook typically grab a location when you open the app. So frequent users of social media are more likely to have a fresh, cached location that non-social media users. If you are targeting business travelers, the cached location might the last city before they got on a plane. Or, it could be your home neighborhood and not where you work or go to games.
  • Wifi turned “on”. If the Wifi is turned on then the device can access the location service and there is a much greater chance that the initial result is fairly accurate. If you didn’t have a chance to read Part 1, when the Wifi is on your browser gathers local Wifi node information from your Wifi card, and it can use that information in a location service provider request over the internet to try and triangulate your position. Typically this means your initial location can be within a block or two of the actual position. Also, it is possible if Wifi is turned on that you can get a significantly more accurate initial location than if you were using GPS by itself with no Wifi or internet.
  • Internet connectivity strength. If you have a poor internet connection and no Wifi, then the browser’s requests to the location service can be delayed, blocked or even interrupted.
  • No VPN. Take note commercial application developers: as mentioned in Part 1, if VPN software is in use it can wildly affect accuracy and even place you in another State (or Country).

Can I use HTML5 Geolocation for mobile tracking? Yes, with caveats. Typically HTML5 tracking applications are built inside a native wrapper framework such as PhoneGap or Titanium. There are several immediate problems with stand-alone, browser-only HTML5 tracking applications. First, there is no built-in functionality to keep the screen from going to sleep. Second, when the screen goes to sleep the HTML5 Geolocation functionality also goes to sleep. Native-based tracking applications can work around these limitations and listen passively in the background when they are minimized. Third, you have little control over the GPS settings to help management battery consumption.

Can I use HTML5 Geolocation offline? Yes! If there is no cellular connection or Wifi available, then HTML5 Geolocation can still access cached locations and real-time GPS information. This is vastly different from what was discussed in Part 1 as related to applications targeted at laptops, desktops and tablets that may or may not have GPS. If a device does not have a built-in or externally available GPS then your offline application will not work.

Handling abnormal location results. Your application will occasionally encounter widely inaccurate results and you need to handle these gracefully for the best user experience possible. My recommendation is to check the timestamps and distance traveled between the current geolocation object and the previous one. If the distance or speed seems excessive then you’ll need to reject the result. In the reference section below is a link to more information on calculating the distance between two points containing latitude and longitude. As an example, see the attached screenshot with the spurious results indicated by red circles. Also note in the screenshot the accuracy level was 3 meters, so it’s important to understand that even at high accuracy levels you still need to very that each location meets your minimum requirements. This way your results will always look polished and professional to the end user.

Spurious results

What are some of the downsides of using HTML5 Geolocation versus native? The bottom line is that for simple location gathering and basic tracking HTML5 Geolocation is just fine. This should meet the requirements for most consumer applications. For anything more complex than that you should consider looking at going native.

  • It may not work on older phones and older browsers (depending on your definition of old). See below in the references section for a link to a fallback library to handle these situations.
  • HMTL5 Geolocation offers significantly less control over GPS settings. This can have an unacceptable impact on more complex applications.  Because of this, I also suggest that HTML5 Geolocation is not suitable for long-running tracking applications.
  • Battery life management. This is a direct result of bullet #2. It’s more challenging to manage battery life with HTML5 Geolocatoin if your requirements call for continuous use of the GPS.  Your control is very limited with respect to these two properties: timeout and maximumAge.
  • Cannot use it when the application is minimized. If your requirements calls for the ability to passively receive locations while in a minimized state then, as mentioned earlier, you will have to go native.
  • Very little control over how often you want location updates. You’ll need to do a bunch of custom coding to emulate what is already built into native application APIs. For example, the native Android API offers very detailed control over what type of geolocation data you can get access to, how you can access it and how often. Read more on that topic in my post on How Accurate is Android GPS Part 1 – Understanding Location Data and also take a look at Android’s LocationManager Class.

References

W3C Geolocation API Specification 

HTML5 Geolocation Test Tool

Mozilla – Using Geolocation

Calculating distance between two points.

Geolocation fallback library for older browsers

How accurate is Android GPS? Part 2 – Consuming real-time locations

In part 1, we looked at the six types of location data available through the Android native SDK and talked about how to work with cached location data. As others have also discovered, there is a ton of information to say on this topic and way more than you could pile into a single blog post. So, this post focuses on working with real-time locations. Once you activate the location listeners they will start sending you information and knowing how to handle that data is what this post is about.

The most important take away I hope to leave you with is take time to understand your accuracy requirements and your user’s basic geographic behavior patterns. I’ve tried to liberally sprinkle example use cases to help illustrate some of the concepts.

Time and distance properties. The first thing you’ll notice when you start building location aware Android apps is you can set time and distance properties that affect how often the device will request a location update. This is accomplished using the overloaded LocationManager.requestLocationUpdates() method. These properties adjust how aggressively the device will request location updates by minimum time elapsed and minimum distance traveled. How you adjust these settings significantly affects the battery life. The equation for battery life is straightforward and simple: the shorter the minimum time interval the faster the battery will be drained.

In the code snippet below, the time and distance properties are both set to zero, which is the most aggressive setting.  You’ll get a location update if any change in location is detected at any time. This setting can result in your app accessing the GPS provider as often as 50 – 60 times or more per minute.

locationManager.requestLocationUpdates(LocationManager.GPS_PROVIDER,0,0,listener);

I’ve experimented with this setting by leaving it running, and it drained a fully charge Samsung Galaxy SIII in less than 3 hours. The phone also got very hot to the touch. It takes considerable power to run the current generation of GPS chips. Even using network location requires that the phone use its radio to make an internet request to the location processing server. So, carefully evaluate your use cases and be conservative about the time and distance properties that are needed.

Except for high performance situations, it will be rare when you need location access every second at a zero distance delta. You can also adjust these properties if your app detects a change in the usage pattern. For example, if the user slows down for a period of time then you might restart the listener using a longer time interval. The listener does not have to stay static and be set only once. Hopefully this is your “ah-hah!” moment. In fact I encourage you to make dynamic adjustments to the intervals as your usage scenarios change. The GPSTester tool allows you to easily experiment with different configurations.

Here are some use case examples:

  • Real-time tracking app for use when walking, running, biking or driving.  As expected, this app will require the most aggressive settings for time and distance.
  • Delivery vehicle tracking app. The delivery vehicle will be starting and stopping all day long. Minimal ability to plug phone into charger. This app should dynamically adjust its time and distance settings.
  • App only needs a one-time snapshot location to find nearby coffee shops. This requires that the app run until a minimum accuracy value is received and then it can turn off the location listeners. Other examples are placing a geotagging a photo, tweet or Facebook posting. You could also try your luck using the overloaded Location.requestSingleUpdate() method.

And, here’s a very basic set of technical requirements for dynamically changing requestLocationUpdates() settings:

  • Start application using minTime = 0 and minDistance = 0. Use an aggressive setting to get accurate location as quickly as possible.
  • Once accuracy is less than 50 meters and speed less than 45 mph set minTime = 5000 and minDistance = 25.
  • Speed equals 0 for greater than 1 hour. Shut off location listeners and notify user.
  • Battery gets low. Shut off location listeners and notify user.

Time to 1st real-time location. It can take up to several minutes to get the first GPS or Network location result. Use effective notifications to let your users know there will be a delay in getting this information. For example, many Android mapping apps use an accuracy circle around the current location indicator to give a sense of accuracy level. 

As you can see in the GPS Provider screenshot taken from the GPS Tester app, using a warm GPS it took ~9 secs to get the first GPS result from the device and then 1450ms to get the next update for an accuracy of 3.0 meters.  By warm I’m referring to a device where the GPS and Network location have been accessed recently.

How long it takes to start getting accurate results depends on many different factors of which you, as the developer, will have very little control over but you’ll need to plan for accordingly. As discussed in Part 1, many developers lean heavily on the cached locations to try and work around these types of delays. Yet, depending on your use cases and the end user personas these results can either be very useful or fantastically inaccurate.

According to Google user interface responsiveness guidelines 100 to 200 milliseconds is where users will start to perceive slowness and we are talking about 9 whole seconds here that the user had to wait. Of course 9 seconds is “fast” for a GPS acquisition. And, you can typically expect much longer time frames, especially on a cold device, and until the accuracy drops down to within a few hundred meters or less.

I’ve seen differences in acquisition times between two phones of the same exact model and same settings even though they were placed right next to each other. Other factors that affect acquisition times can be as simple as where you placed the phone in the car such as down by the gear shift between the seats or up on the dashboard. Or the user could be standing just inside a restaurant under a metal awning because it’s raining outside.  The list goes on and on. Cement, bricks, metal, car bodies, heavy foliage and buildings are some other examples of things that can interfere with GPS signal accuracy and the time to first acquisition. Like the Boy Scouts say, “Be Prepared” and be kind to your users by using user interface notifications to let them know of location acquisition delays.

Streaming real-time locations. Once the device starts providing real-time locations, my suggestion is to check the accuracy of each result and consider rejecting those greater than a certain amount that are based on your requirements. The longer the location provider runs, and if the device has an unobstructed views of the sky and good cellular connection, then typically the accuracy will improve up to a certain point and then level off, and then it will fluctuate.  Here’s a pseudo-code snippet showing how to check the accuracy of each GPS location result:

public void onLocationChanged(Location location) {
     if(location.getAccuracy() < 100.0 && location.getSpeed() < 6.95){
          //Do something
     }
     else{
          //Continue listening for a more accurate location
     }
}

Here are some rough examples of accuracy thresholds I used for a project last year. Your requirements may vary as to how these different thresholds will affect the behavior of your application; these were examples that required geocoding that converted the current location to an approximate address. Depending on the result the application gave different feedback to the user:

  • Rooftop  <= 10 meters (desired result)
  • Street >10 meters and <= 100 meters (let user know it’s close but not perfect. Good enough?)
  • Neighborhood > 100 meters and  <= 500 meters (give visual feedback that accuracy is low)
  • City > 500 meters and <= 2000 meters (ask user to verify city name from a list)
  • County > 2000 meters (prompt for manual location input)

Take into account your own unique use cases. You might completely reject any accuracy value greater than 100 meters (328 ft) if your app simply helps people find open parking lots at NFL games. You could have an app that returns a list of Dentist offices within a 5 mile (8000m) radius. Or a weather app could only need to know approximately what city you are in. These are just ideas to help get you thinking.

Not all real-time location data is alike.  Continuing on the theme of consuming real-time location, let’s dig into a few more examples of why it should be looked at closely. Here’s one thing that I keep forgetting even though I’ve blogged about it: you can get null location values and if not properly handled they will crash the app. That’s an easy one to forget until your users start to report random application crashes.

You can also look for what I’ll call spurious results. These are results that are way outside what you might consider a running average. Check the distance between the last location result and the current location result using the Location.distanceBetween() method. If the distance and speed required to cover that distance is significantly greater than the running average, then you can reject that result.

I also want to mention that simply holding the device in one location while standing outdoors can result in the latitude/longitude wandering over a reasonable distance. Even if you set the phone on a rock the results can wander. Just keep this in mind if you have accuracy requirements. Just because the user is standing still outdoors doesn’t mean that the indicated location will stay exactly the same. It could wander over a 10 to 50 meter radius or more.

I have occasionally seen wild location fluctuations that were enough to make a mapping application nauseatingly jump back and forth between different center points. Unfortunately, when it happened I didn’t have the device on a debugger, but I suspect it had to do with the phone detecting various WiFi end points when I stopped at stoplights along my route. It’s possible the phone tried to resolve those WiFi locations using the Google Network Location service and that there was some lag time in processing those results.

Are Android, off-the-shelf, retail smartphones as accurate as something like a Trimble Pro Series Receiver? No way, not all, definitely not! If you have high accuracy, field-usage requirements calling for sub-meter results you should not be using a typical retail Android device. For example, if you are standing in an intersection with four manhole covers that are two feet apart and form a circle, you could not accurately map which manhole cover is the right one that you need to be working in if you are using an off-the-shelf Android. Using my own phone as an example, on a good day my Samsung Galaxy SIII occasionally has down to 3 meters GPS accuracy for short periods of time and then it can start wandering.

Comparing Network and GPS locations and using Criteria. Yes, you should definitely compare the results between these two. The more data you have the better. There’s not a whole lot to say about this other than look at timestamps, accuracy and distance traveled factors. In my experience, just something to keep in mind is the network locations happen much slower than the GPS. I expected this because of the lag time involved with the phone sending information to the remote location service and then waiting for a result to come back. You can read more about how location services work in this article.

The use of android.location.Criteria can also be used to control which listener(s) are used. You can experiment with Criteria using the GPSTester tool. The only minor caveat is it doesn’t include all possible criteria in the current version (v1.2.1.1). My general recommendation is to skip Criteria and hard code in the validation rules. I don’t currently subscribe to the idea that the Criteria will know what’s best for my end users. Please comment below if you feel otherwise, but in my own experience I haven’t come across a use case where Criteria gave what I thought was the right answer. In one use case, the Criteria was set to not incur costs, yet the value returned was to use the network provider even though GPS was available. Using the network location service would have incurred bandwidth costs.

What to do when app is minimized? A common workflow is for a user to minimize the app and then either forget about it or come back to it later. I’ve included this because it’s common for the user to get a phone call or a text message that results in the app being minimized. In most of the use cases I work with, the requirements call for both the GPS and Network Location services to be shut off when the app is minimized. It’s just way too easy for the user to forget about the app and then it could kill the battery in a very short time.

There is the option of using a passive listener. I rarely use these because it assumes some other unknown application will spin up the GPS or Network Location providers.  Your use cases will help you decide whether or not implementing a passive listener is a good idea. If your target audience is a 20-something student who is constantly using location based social media such as Facebook every hour and eighteen hours a day then there’s a good chance the passive listener will return recent and mostly accurate results. However, if you are building a commercial-grade application on a work-related phone then a passive listener is significantly less likely, or even highly unlikely, to speed up the applications ability to get a fast fix on startup.

Use Intents when implementing a passive listener, this involves adding directives in the AndroidManifest.xml as well as writing some code. A great example of how to do this can be found here, so I’m not going to reinvent the wheel. Just don’t forget to add the receiver tag to your manifest:

 <receiver android:name=".receivers.PassiveLocationChangeReceiver"/>

When the app restarts you can use information from the passive listener and consider whether or not it meets your accuracy criteria. You can also use the passive listener to run background processes within your app while it is minimized. You’ll use the Location timestamp once the app is opened again. If the timestamp is very recent, such as less than one hour old, then you could consider using the passive location immediately. Otherwise, you’ll be back to waiting for a real-time location result. For this approach consider comparing Location.getRealtimeNanos() to SystemClock.elapsedRealtimeNanos(). If your target audience are business travelers, then for example you might want to reject cached results that are greater than 8 hours old because chances are the user has hopped on a plane.

Privacy, data storage and data consumption. Yeh, it’s a bummer I even had to bring this up. I’m not a lawyer, but every lawyer I’ve worked with on these types of production applications strongly reminds me to have a Privacy Policy and to be transparent about how the data is stored and how the data is used. Don’t try to do this on your own, hire a lawyer or the equivalent of a lawyer in your country, as the case law is constantly changing and can vary from State-to-State and Country-to-Country.

And, that’s the wrap-up for Part 2. Stay tuned for additional posts on this series that cover using the GPSTester tool and how to interpret the results.

References

Google Developer docs – Location Strategies

Android blog – Deep dive into location

GPS Testing Tool (open source)

HTML5 Geolocation API – How accurate is it, really?

A whole new world for app developers – it’s raining cats and browsers!

Holy smokes – how do we support all these different browser types and versions? I’ve been thinking about this for a while, but after attending the AdobeMAX conference this week I got to talk with other developers facing the same issues.  The web app world used to fairly simple: develop for the one or two major browsers running on desktop computers and you were done. We still grumbled a lot and had cross-browser problems that at the time were really annoying. And now we have to deal with not only desktop computers and laptops, but there are also dozens of different mobile browsers and hundreds of new tablet style devices. To make things even more interesting the pace at which new versions of browsers and mobile operating systems are being released is at an unprecedented (and perhaps unsustainable) rate.

What to do, what to do, what to do?  Well if we step back and look at this holistically, there are a few things to consider that can help lay the ground work for building apps in these crazy times. That will help you make better technical business decisions.

For web apps:

  1. What browsers and browser versions your customers are using? How often do you analyze this? If you don’t use a monitoring/analytics tool for this you can always download a free tool to analyze your web server logs. I’ve used Google Analytics in past, but there are plenty of other choices.
  2. What percentages of your visitors are using a particular browser version? Break down your stats again by the browser version. If you still have a large percentage of customers using IE 6 then you might want to consider continuing to support it.
  3. What percentage of your customers use mobile versus non-mobile?
  4. What percentage of customers are from your country? If most of your customers from outside your country then you’ll also have to include localization code.
  5. What type of device is each web site visitor using? Mobile devices can have vastly different screen resolutions and sizes.
  6. Which browsers represent visitors/customers that make you the most money? This is a very important number to know and getting this information can be a bit more involved than the other questions. If you mess up support for these group(s) you’ll really cause yourself business problems.

For native apps:

  1. Have you reviewed your marketplace stats page? One example is the Android Market offers application stats such as platform versions, devices, countries of origin, languages, etc.

What if you haven’t launched your app or website yet? If this is the case, then check out some of public tracking sites for hints. Some good places to start looking are: w3schools and Wikipedia which lets you drill down into different browser versions.  For mobile apps, most IDEs let you set up various layout scenarios that relate to rough categories of dpi (dots per inch). For example, here is an article from Adobe on Flex mobile development and handling different mobile device screen resolutions.

How do I interpret this information? In general, you want to have support for browsers and operating systems that the majority of your customers use. One example is in the retail/consumer industry you absolutely have to support what your customers use. If the vast majority of your customers use Internet Explorer then you will need to support IE or you will lose customers and money. If the vast majority of your visitors use Android, but your biggest spenders use iPhones then take that into your development considerations. If your website or application performs poorly for a small percentage of very high-spending customers then it could cost you your business. If you are lucky enough to have commercial customers consuming an internal-only application, or if you have an ultra-niche audience that only uses a single browser or smart phone operating system then you already have your answer. You build to suit their needs.

Any further advice? Test, test, test and then test some more on real devices and different versions of browsers. Emulators and simulators are not an adequate substitute for real-world checking. If you are a small shop, enlist family, friends and neighbors to test. Have them check out our app or website for a few minutes while they relax in the evening. You can also buy after market handsets on eBay and other places, just be careful that if a deal is too good to be true you probably shouldn’t purchase it. The reality is there are so many devices nowadays that you simply can’t test them all. So, pay special attention to feedback in your market place listing or on your forums.