GPS Sensor Fusion Toolkit

Getting started

Requirements

Unity 2021.3 LTS or newer. No other packages required.

Installation

Three ways to add the package to your project:

Local UPM package (recommended) — Window > Package Manager > + > Add package from disk… and select the package’s package.json.
Embedded package — copy the com.kephstudio.gps-sensor-fusion folder into your project’s Packages/ folder.
Classic import — copy the Runtime/ scripts into Assets/ (you lose assembly-definition isolation).

Overview

GPS Sensor Fusion fuses GPS (low rate, position + velocity) with an optional AHRS/IMU (high rate, heading + turn rate) into a smooth, high-rate estimate of position, velocity and heading. Two filters are provided: a 6-state linear Kalman filter, and an 8-state Extended Kalman Filter (CTRV) that adds heading and turn rate and ingests AHRS measurements between GPS fixes.

What’s included

KalmanFilter	6-state linear filter (constant velocity) — simple and fast, GPS only.
ExtendedKalmanFilter	8-state CTRV filter — best quality, fuses GPS + AHRS through turns.
HeadingFilter	Adaptive heading smoother (EMA + rate limit + stationary dead-zone).
GPSMovementState	CTRV dead-reckoning predictor between fixes.
NMEAGenerator / NmeaParser	Standard NMEA 0183 generation and parsing into a `GpsFix`.
GeoUtils	Geodesy: haversine, bearing, unit conversions, lat/lon → local ENU.
IGpsSource	Pluggable input: NMEA push, Unity LocationService, CSV replay.
SfMatrix	Zero-allocation dense matrix backing the filters.

Coordinate conventions

All filters operate in Unity world meters: X = East, Y = Up, Z = North. Heading is in radians, 0 = North (+Z), increasing clockwise. Convert GPS latitude/longitude to a local metric frame around an origin before feeding the filter — GeoUtils.GpsCoordTo3D(refLat, refLon, refAlt, lat, lon, alt).

Quick start

using UnityEngine;
using KephStudio.SensorFusion;

public class Example : MonoBehaviour
{
    private ExtendedKalmanFilter _ekf;

    void Start()
    {
        // processNoise, measurementNoise, headingNoise, turnRateNoise
        _ekf = new ExtendedKalmanFilter(1f, 3f, 0.1f, 0.1f);
        _ekf.Reset(transform.position, Vector3.zero, heading: 0f, turnRate: 0f);
    }

    public void OnImu(float headingRad, float turnRate, float dt)   // ~50 Hz
    {
        _ekf.Predict(dt);
        _ekf.UpdateAHRS(headingRad, turnRate, 0.05f, 0.05f);
    }

    public void OnGps(Vector3 worldPos, Vector3 worldVel)           // ~10 Hz
        => _ekf.UpdateGPS(worldPos, worldVel, 1f, 1f);

    void Update()
    {
        transform.position = _ekf.GetPosition();
        transform.rotation = Quaternion.Euler(0f, _ekf.GetHeading() * Mathf.Rad2Deg, 0f);
    }
}

Call Predict once per elapsed dt, then apply whichever measurement arrived — GPS and AHRS updates are independent and run at their own rates.

Tuning

Estimate lags real motion	Lower `measurementNoise`, or raise `processNoise`.
Estimate is jittery	Raise `measurementNoise`, or lower `processNoise`.
Overshoots in turns (EKF)	Raise `turnRateProcessNoise` and feed AHRS turn rate.
Poor fixes should count less	Pass larger position/velocity uncertainty (e.g. scale by HDOP).

Samples

Import samples from the Package Manager (Samples tab).

Synthetic Track Demo — Noisy GPS track → EKF → plots truth / noisy / filtered with LineRenderers. No hardware.

Support & license

Distributed under the Unity Asset Store EULA. Questions, bugs and feature requests are handled directly by the studio — contact@kephstudio.ch.