Kalman Location Engine

[cammace]

In order to better control when location updates occur, a Kalman filter can be used to force location updates between real GPS updates.

cc: @zugaldia

[danesfeder]

With 0.20.0, Location filtering is pushed to Navigator and will no longer be on the client as a responsibility, closing

[Ohemgi]

hey @danesfeder!

Location filtering is pushed to Navigator and will no longer be on the client as a responsibility

Do i understand you right, that a kalman filter is not required client side / on mobile device? I'm looking for smooth location transitions and a proper "location estimation" while in navigation mode and locationLayer.forceLocationUpdate(location) in combination with the locationEngine does not provide "enough" accurate updates. The location marker appears to "lag" and the position is always "behind".

If this is the case, can you suggest a different approach for that problem?

[danesfeder]

Do i understand you right, that a kalman filter is not required client side / on mobile device?

Hey @Ohemgi 👋 you are correct here - the SDK will ingest the raw locations and filter / snap them (returning in ProgressChangeListener).

The location marker appears to "lag" and the position is always "behind".

We have actually been working on better prediction / projection of the snapped locations returned. Do you mind providing which version of the SDK you've been testing? Thanks!

[Ohemgi]

Hey @danesfeder Thanks for the reply on this. Im currently on 0.19.0

[danesfeder]

@Ohemgi these improvements were made in 0.21.0 - would you mind updating? We made more improvements in 0.22.0 slated for a release this week.

[Ohemgi]

@danesfeder ill check that out and comment later on the improvements. Thanks for the hint!

[danesfeder]

@Ohemgi specifically #1354 is what I'm referring to - you can adjust that value (NAVIGATION_LOCATION_ENGINE_INTERVAL_LAG) in the MapboxNavigationOptions, but please test the default before doing so.

[Ohemgi]

@danesfeder sorry for putting this request here, but since you are tracking the sdk changes i think you might be the correct one beeing asked 😈 :
After updating to 0.21.0 the location prediction / projection seems to work pretty nice and accurate. Good job on that so far 👍

But one "blocking" thing came up for me: Without any further changes (except the dep update) the routeProgress object, raised by the navigation in ProgressChangeListener, holds the relevant progress data as before, except the distance information. Every distance field is 0.0 (like routeProgress.distanceRemaining). Since the navigation setup / input parameters remained the same, you might could give me a hint which part of the sdk i should dive into, to debug this? Just to clarify: There is no exception etc. raised. onProgressChange Event is still raised but with the Zero-Distance-Values as described above.

Thanks in advance, Ben

[zugaldia]

@Ohemgi could you cut a separate ticket for this new question? Looks like a regression worth looking into.

[danesfeder]

Hey @Ohemgi have started progressing along the route when you're seeing this behavior? How are you testing the SDK in this scenario?

In 0.21.0 we saw the the SDK was returning 0 values until progress was made along the route and we had sufficient location updates from the device to be considered "tracking". Related ticket #1404.

Now when the route starts, we should provide full route duration / distance values rather than 0. This will be fixed in our 0.22.0 release slated for this week. In the meantime you can test with a SNAPSHOT that will have these updates #1404.

If this doesn't fix your issue, as @zugaldia mentioned, please open a ticket and we will investigate further. Thanks!

[Ohemgi]

Hey @danesfeder ,
Hey @zugaldia ,

ok, i was already about creating an issue, but your suggestion sounds like it could directly adress my problem @danesfeder 👍 😄 I'll check the snapshot build and comment on the results after doing so.

Thanks for your participation on that behavior!

Edit:

when you're seeing this behavior? How are you testing the SDK in this scenario?

I'm testing it on an android 7 device with mocked locations. I see this behavior on startup. Since i check for certain distance-progressions, they fail due to the zero-values

Br, Ben

[Guardiola31337]

Hey @Ohemgi 👋 friendly reminder that 0.22.0 version of the SDK was released yesterday 🎉

As suggested above, please feel free to cut a new ticket if you still experience the issue. Thanks a lot!

[Ohemgi]

Hey @Guardiola31337 👋 nice! thanks for the update on this. I updated to 0.22.0-SNAPSHOT yesterday. Sadly the first progress instance still holding zero-values but i gonna check my setup and give a feedback tomorrow 👍

[Guardiola31337]

Sweet! Thanks for the update @Ohemgi 🙏