18

I am trying to make sense of an inexplicable situation — starting February 15, a remarkably high percentage of the location readings we captured in our app are reporting exactly 10.0m accuracy, which seems strongly correlated with those arriving from a GPS source, as we see it almost exclusively with High-Accuracy and Sensor-Only mode (though there have been occasions where Battery Saving).

After reviewing our system, we don't see any point where we could be introducing this ourselves. It affects all deployed versions of our app.

We have begun to instrument our application to get more metadata, but would love to have a better understanding of how something like this could happen approximately instantaneously, which affected multiple versions of the app at a time when we did not change any running services.

Here area a few potentially-related issues:

And some graphs to show what we are seeing:

Distribution of accuracy before/after Feb 15

Probability of 10m accuracy by location mode

So — has anyone else seen anything similar, or have an explanation as to what might be going on here?

Rann Lifshitz
  • 4,040
  • 4
  • 22
  • 42
Eric O'Connell
  • 858
  • 5
  • 14
  • We are experiencing the same, we are the developers who posted this question https://stackoverflow.com/questions/49348489/cordova-geolocation-accuracy-gets-capped-at-10-meters – Paranoid Android Apr 09 '18 at 10:33
  • YES. We also started getting reports of accuracy changes around then including on devices that did not get Play Services updates. I posted https://stackoverflow.com/questions/49289206/why-does-fusedlocationproviderapi-never-report-accuracy-better-than-10m-is-this – Hélène Martin Apr 09 '18 at 20:32
  • The fact that so many people are seeing this, and across both iOS and Android, is increasingly leading me to the strange belief that this has something to do with the GPS system itself. I still feel very hesitant about that, but I'm beginning to run out of hypotheses that support the available data. – Eric O'Connell Apr 09 '18 at 22:44
  • @EricO'Connell Try the test app I linked to in my post -- https://github.com/will-quast/android-location-demo. You should see that using the old non-Play-Services APIs still gives high accuracy. What I'm thinking is that Play Services APIs phone home at some point and picked up some kind of policy change. – Hélène Martin Apr 10 '18 at 00:13
  • I am seeing the same using this plugin https://github.com/mauron85/cordova-plugin-background-geolocation/tree/2.x, very good results. – Paranoid Android Apr 23 '18 at 08:40
  • Something fishy is definitely happening here. We have an app that's been around for 6+ years, and one of its main features is to measure from one point to another while you walk with your phone. Accuracy is important here, and a couple months ago our users started reporting they can't get better than 10m accuracy. On the same app, we used to regularly see 3-4m accuracy with good conditions. I've confirmed this is the case on MANY Android phones. It seems like something changed that's out of our control, and it's making users not trust our app. Any help or updates are much appreciated. – DiscDev May 01 '18 at 20:52
  • Has anybody submitted a bug report to Google? If so please post the link here so we can all star it. – DiscDev May 01 '18 at 21:53
  • https://issuetracker.google.com/issues/79189573 – paul Jul 10 '18 at 17:29
  • It's October and the issue is still there! – Paranoid Android Oct 01 '18 at 12:32

2 Answers2

1

it is a Google Play Service issue, reported internally here and it will be fixed from version 13.4.0

Paranoid Android
  • 4,672
  • 11
  • 54
  • 73
0

I am not a GPS expert, but years ago I picked up somewhere that GPS technology is (still) a military technology, meaning that if military purposes require it, civil usage/accuracy of GPS functionality may be limited from time to time...

But this was years ago. Of course (and hopefully) I am mistaken, but just wanted to share this possibility with you.

Edit:

Never mind. After reading this, I understand that this "Selective Availability" feature is abandoned a long time ago.

Bart Hofland
  • 3,700
  • 1
  • 13
  • 22