kalman_dr

Your app goes blank when GPS drops out. kalman_dr keeps the position alive.

Use it alongside geolocator — when GPS fails in tunnels, urban canyons, or parking garages, kalman_dr maintains your position estimate using device sensors until GPS returns.

4D Extended Kalman Filter with covariance-driven accuracy reporting. Pure Dart, no native dependencies.

Features

4D state vector: latitude, longitude, speed, heading
Covariance-driven accuracy: honestly degrades over time during GPS loss
Safety cap: stops at 500m accuracy — no false confidence
Two modes: EKF (full) and linear extrapolation (lightweight)
Decorator pattern: wraps any LocationProvider transparently

Install

dependencies:
  kalman_dr: ^0.3.0

Quick Start

import 'package:kalman_dr/kalman_dr.dart';

// Create a filter with an initial GPS fix
final filter = KalmanFilter.withState(
  latitude: 35.1709,
  longitude: 136.8815,
  speed: 12.5,
  heading: 90.0,
  timestamp: DateTime.now(),
  initialAccuracy: 5.0,
);

// Predict position forward by 1 second (GPS lost)
final predicted = filter.predict(const Duration(seconds: 1));
print('${predicted.lat}, ${predicted.lon} '
    '(accuracy: ${predicted.accuracy.toStringAsFixed(0)}m)');

// Update when GPS returns
filter.update(
  lat: 35.1710,
  lon: 136.8820,
  speed: 12.8,
  heading: 91.0,
  accuracy: 4.5,
  timestamp: DateTime.now(),
);

Wrap a location provider

final provider = DeadReckoningProvider(
  inner: SimulatedLocationProvider(),
  mode: DeadReckoningMode.kalman,
);

provider.positions.listen((position) {
  // Receives GPS when available, Kalman predictions when GPS is lost
  print('${position.latitude}, ${position.longitude} '
      '(accuracy: ${position.accuracyMetres}m)');
});

Integration Pattern

The package becomes most useful when it sits between your raw location source and the rest of the Flutter app. The pattern is: create the underlying GPS provider once, wrap it with DeadReckoningProvider, then surface the stream in UI code that can explain when the app is running on prediction instead of live GPS.

import 'package:flutter/material.dart';
import 'package:kalman_dr/kalman_dr.dart';

class DeadReckoningStatusCard extends StatefulWidget {
  const DeadReckoningStatusCard({
    super.key,
    required this.gpsProvider,
  });

  final LocationProvider gpsProvider;

  @override
  State<DeadReckoningStatusCard> createState() =>
      _DeadReckoningStatusCardState();
}

class _DeadReckoningStatusCardState extends State<DeadReckoningStatusCard> {
  late final DeadReckoningProvider provider;

  @override
  void initState() {
    super.initState();
    provider = DeadReckoningProvider(
      inner: widget.gpsProvider,
      mode: DeadReckoningMode.kalman,
    );
    provider.start();
  }

  @override
  void dispose() {
    provider.dispose();
    super.dispose();
  }

  @override
  Widget build(BuildContext context) {
    return StreamBuilder<GeoPosition>(
      stream: provider.positions,
      builder: (context, snapshot) {
        final position = snapshot.data;
        if (position == null) {
          return const Text('Waiting for location...');
        }

        final degraded = position.accuracyMetres > 25;
        return ListTile(
          title: Text(
            '${position.latitude.toStringAsFixed(5)}, '
            '${position.longitude.toStringAsFixed(5)}',
          ),
          subtitle: Text(
            degraded
                ? 'Predicted path — accuracy '
                    '${position.accuracyMetres.toStringAsFixed(0)}m'
                : 'Live GPS lock — accuracy '
                    '${position.accuracyMetres.toStringAsFixed(0)}m',
          ),
        );
      },
    );
  }
}

This is the tunnel pattern: keep the location pipeline alive, surface the degraded confidence honestly, and let the rest of the map/navigation stack keep rendering instead of freezing.

API Overview

Type	Purpose
`KalmanFilter`	Predicts and updates the 4D state vector for dead reckoning.
`DeadReckoningProvider`	Wraps a location provider and emits predicted positions during GPS loss.
`DeadReckoningMode`	Selects EKF or linear extrapolation mode.
`KalmanPosition`	Carries predicted position, speed, heading, timestamp, and accuracy.

Safety

Display-only position estimates — does not control vehicle systems. When accuracy exceeds 500m, the provider stops emitting rather than showing unreliable data. Built with automotive-grade test discipline (200+ unit tests), usable in any Flutter app.

Works With

Package	How
geolocator	Feed geolocator's position stream into `DeadReckoningProvider`
flutter_map	Render predicted positions on the map during GPS loss
latlong2	Shared coordinate types

License

BSD-3-Clause — see LICENSE.