simondlevy / tinyekf

Lightweight C/C++ Extended Kalman Filter with Python for prototyping

TinyEKF: Lightweight C/C++ Extended Kalman Filter with Python for prototyping

TinyEKF is a simple, header-only C/C++ implementation of the Extended Kalman Filter that is general enough to use on different projects. It supports both single- and double-precision floating-point computation. In order to make it practical for running on Arduino, STM32, and other microcontrollers, it uses static (compile-time) memory allocation (no "new" or "malloc"). The examples folder includes both a "pure C" example from the literature, as well as an Arduino example of sensor fusion. The python folder includes a Python class that you can use to prototype your EKF before implementing it in C or C++.

Arduino users can simply install or drag the whole TinyEKF folder into their Arduino libraries folder. The examples/SensorFusion folder contains a little sensor fusion example using a BMP180 barometer and LM35 temperature sensor. I have run this example on an Arduino Uno and a Teensy 3.2. The BMP180, being an I^2C sensor, should be connected to pins 4 (SDA) and 5 (SCL) of the Uno, or pins 18 (SDA) and 19 (SCL) of the Teensy. For other Arduino boards, consult the documentation on the Wire library. The analog output from the LM35 should go to the A0 pin of your Arduino or Teensy.

In addition to the class definition, the python folder has an example of mouse tracking, using OpenCV. So you will have to install OpenCV to run this example. There is also a sensor-fusion example in this folder.

tinyekf's Issues

SensorFusion Example

Is Line 50 correct ???

        hx[2] = this->x[1]; // LM35 temperature from previous state

Should it be (index 2 for LM35) ???:

        hx[2] = this->x[2]; // LM35 temperature from previous state

How do you adjust the K "Kalman" value

How do you adjust the K "Kalman" value

Use library - get matrix

Hi,
I'm looking to correct the position given by my GPS (NEO-M8T) with the use of an IMU.
How can I get the matrix to use your algorithm?

I tried your example with the temperature readings and worked pretty well, But now I want to fuse two different things.

Best Regards

Have no SFE_BMP180.h

When i use the code and have the error :no such decirtory SFE_BMP180.h,i don't know why .I just have imu and magnetometer.Looking forward your reply!

Predicting translation xyz values?

Hi, thank you for making this code available. I have some sensor measurements that come in at 30Hz. I need to 'upsample' this data to 60Hz, so am looking at the kalman filter to predict the measurements in between the samples. Is your filter suitable for this? If you have a minute, would you be able to point me in the right direction to do this using your code? Thank you!

C++ TinyEkf::predict() should support actually adding process noise

possible bug? shouldn't that x[1] be x[2]?

TinyEKF/examples/SensorFusion/SensorFusion.ino

Line 54 in 1a5d606

hx[2] = this->x[1]; // LM35 temperature from previous state

Cheers!

Examples ekf.step(z) call does return an failure value (=1) - Why?

Hello I tried to get the Arduino examples working. Started with the Benchmark example - does run OK. Then the SensorFusion example. I have adapted this to the BME280 and mentioned than every cycle ekf.step(z) does return an '1' value which is a failure as read in tiny_ekf.h "caused by non-positive-definite matrix" from (trapped in choldc1()).
I have checked all, but couldn't find any issue. Then I had a look to the Benchmark example. The result is the same.
The estimated x values vector does reach their target value after some steps.
I'm not sure what does failure really mean. Is this correct? Is it a real problem - or an bad model?

When H is non square I get "ValueError: operands could not be broadcast..."

On the python version I get this error ValueError: operands could not be broadcast together with shapes when I have a H matrix which is not square, in particular when I have more states than sensors so when n>m.

The error is raised by this line
https://github.com/simondlevy/TinyEKF/blob/master/extras/python/tinyekf/__init__.py#L59

Probably it's the same problem reported in #5

broken link: http://home.wlu.edu/~levys/kalman_tutorial/

Thanks for sharing!!! :)

ValueError: operands could not be broadcast together with shapes....

Asynchronous sensor readings

Great tutorial and codebase Simon. It cleared up quite a bit of the concepts behind an EKF.

I am seeing some limitations with TinyEKF and I would like to know if this is right:

• The filter cannot handle null data. If new measurements are received for only one sensor, we can either filter using old data for the other sensor (making it appear as if it was new data that just happened to be exactly the same), or wait until we have new data for each sensor.

• The filter cannot handle variable timesteps. The measurement function is not based on any timestamps and I assume needs to be hardcoded for a specific timestep.

To be clearer, I was planning on using this filter to fuse orientation data coming from 2 sensors at different rates. I can deal with a fixed timestep, but reusing old data for the slower sensor is problematic, as well as waiting for it to update.

Delta t Variable in Model Method

I may be reading this wrong, but I don't see a delta t passed to the model method. I'm trying to find the position using the previous velocity is integrated with respect to dt: Fx[0] = this->x[0] + this->x[1]*dt. In this case position is the zero index and velocity is the one index.

I must be missing something, am I supposed to pass delta t via the state vector?

Matrix dimensions in error covariance update step incorrect?

On this line, the vector z-hx of size m is multiplied by the matrix G, of size n x m:
https://github.com/simondlevy/TinyEKF/blob/master/src/tiny_ekf.c#L321

The result of which is a vector of size n, not the m x n matrix tmp2 that the result is written into.

Now I'm not using the provided matrix functions so I can't say if this works or not, but regardless it's confusing when reading the code and could introduce issues. There should be a tmp6 array of size n that the multiplication of G with z-hx is written into, and used in the subsequent addition with fx to obtain x.

ImportError: No module named realtime_plot

Thank you for this useful piece of software!
Please make the RealtimePlotter dependency more evident 😃

Help needed in velocity estimation

Hi,
I am new to kalman filters and this library has proved very useful to me as a stepping stone to learn more about them. I wish to implement the estimation of velocity using this library. I have one sensor giving out only velocity values and another sensor giving me acceleration values. So my model would be-
V(k)=V(k-1)+a.t
A(k)= A(k-1)
However i am unsure where to put this. Does this go in the process model or my measurement function? I am confused and would be grateful if you could help. Thanks.

What is the difference from kalman？

Many thanks to the author for contributing the code. But is there any difference between EKF and kalman. No jaccsd, no Taylor?

Does the Jacobian get updated?

Thank you for making the awesome tutorial and code available.

Question in the python sample code.

https://github.com/simondlevy/TinyEKF/blob/master/python/tinyekf/__init__.py

Why getH(x) and getF(x) are only called once during the init? Shouldn't them be updated each step()?

TinyEKF For GPS filtering?

Is that possible to use this library for GPS Coordinates, I'm getting GPS latitude and longitude from my gps module, is that possible to filtered out the noise using this library?

C#

Any chance of a C# implementation?