gnss imu fusion

Liuyanting

I found a library about improving GPS stability Can it be added to racechrono to improve track record quality? Or maybe something like the u-blox Dead reckoning series like the Neo-M9V Improved position report update rate through sensor fusion, from 5hz to 50hz In this way, can racechrono achieve 10hz or higher update rate using the phone's built-in GPS and sensor fusion?

This is the library link https://gps.maddevs.io/en https://maddevs.io/blog/reduce-gps-data-error-on-android-with-kalman-filter-and-accelerometer/ https://github.com/maddevsio/mad-location-manager

aol

Interesting, thank you for the tip. I will take a look at it.

krisa9977

Interesting topic. I’m actually surprised nobody seems to have implemented this properly yet.

GNSS + IMU fusion looks especially useful for lap timing when GPS reception becomes poor or drops out completely for a few seconds. In that situation, using the last known position/velocity together with the accelerometer and gyro should be enough to estimate the car’s movement until GNSS comes back.

For circuit driving the problem is even easier, because the approximate track geometry is already known.

It would be interesting to know if RaceChrono currently does anything like this, or if lap timing still relies entirely on GNSS positions and interpolation between valid fixes.

aol

krisa9977 RaceChrono does lap timing entirely based on GNSS.

We looked into fusing IMU data with GNSS to improve accuracy, but it never went anywhere. Doing it properly would require far more expensive IMUs, rigid mounting to the vehicle, and proper calibration. And devices like RaceBox and Dragy already deliver GNSS accuracy good enough that the added effort wouldn't pay off.

Cali_Chuck

You're expecting too much from a device that isn't designed for your application.

Just buy a Racebox, it does all this for you with extremely high accuracy and 25Hz update rate. It also has a purpose built antenna and multi-constellation support. Mounting with a clear field of view is also critical.

The reason you have GPS dropouts is almost certainly the crappy antennas in all modern phones. Most cell antennas actually have negative gain in most points on the sphere surrounding the device. (I work in radio including with GNSS systems.) And also combined with poor placement, this results in signal obstruction by the roof when driving in some directions. Remember that GPS satellite orbits are not polar but rather at 55 deg inclination. So there are most certainly driving geometries where a phone on a dash mount will have a suboptimal geometry relative to whichever birds are presently in "view".

Trying to "fuse" other sensor data just reinforces the point that phones are not designed for high-Hz, high-precision location applications. They are consumer devices engineered for consumer use cases.

My phone has enough to do running RaceChrono and processing BT telemetry streams from OBD/CAN. Even if I could mount it somewhere it could get consistent GNSS receive power I'd rather delegate that function to a fit-for-purpose device. Decoupling lap datalogging from position also lets me put the RaceBox in an idealized location for sky FOV.

krisa9977

Cali_Chuck RaceBox is obviously a great solution, but that’s not really what I’m trying to solve. 🙂
The interesting question for me was whether a phone alone could do it. Turns out the answer seems to be yes.

krisa9977

aol That’s interesting, because I came to almost the opposite result in testing.
I’m getting a smooth 10 Hz trajectory and lap timing using only the phone, and it can continue even with GNSS completely unavailable. No external receiver or expensive IMU.
I’ll post some traces — I’d be curious what you think.

I do agree about rigid mounting though. But if you want useful telemetry from the phone, you need to mount it rigidly anyway. Otherwise lateral/longitudinal G traces are pretty much useless. So I don’t really see that as an additional limitation — only if you’re using the phone purely as a lap timer.

aol

@krisa9977 yes, sure I'm interested in your results!

krisa9977

aol Almost there. I’m just finishing a few last details and then I’ll post some results.

krisa9977

Here’s the quick comparison I promised. Both screenshots show the same area at the same scale:
Phone GPS only:
https://racemeasureapp.com/assets/screenshots/racechrono/racechrono-gps-only.jpg
Phone GPS + IMU fusion:
https://racemeasureapp.com/assets/screenshots/racechrono/racechrono-gps-imu.jpg
With fusion enabled, the trajectory continues between GNSS updates and follows the car’s movement more closely. This is still phone-only, with no external GNSS receiver. I’m not suggesting it replaces a good 25 Hz receiver, but it shows that IMU fusion can be genuinely useful rather than purely theoretical.

aol

krisa9977 Looks good indeed! Can you give me pointers which algorithms you used etc?

krisa9977

aol Sure. It’s a custom 2D error-state EKF: GNSS corrects position/velocity, while accelerometer and gyro propagate a vehicle-constrained model between fixes. Delayed fixes are replayed through a short fixed-lag buffer, with innovation-based outlier rejection. No external library.