Sensor Fusion Software with Matlab Models

[jasonhaedt]

Hello,

I am super excited about the sensor fusion library.  I need to have a means to model your sensors in Matlab.  I see some Matlab scripts but didn't see a real time way to model my sensor fusion algorithms and was curious what Freescale has to enable this type of development with Matlab.

Thank you for your help,

Jason Haedt

[michaelestanley]

Jason,

I don't have models of the fusion library to offer today.  The history is that we developed our original algorithms in Matlab, and translated to C.  Since that time (several years ago), we've made all our changes directly in the C code.  The reason being that with our visualization tools, we can generally replicate hours of Matlab work with a few hand gestures.  So our code has diverged significantly from those original models way back.

We do have one additional capability which we haven't yet put into the public domain, but I've been considering it.    Back when we were developing our algorithm in Matlab, we needed a way to define "objective truth" in terms of expected response of the algorithm.  I built a Matlab library to model device trajectories and expected sensor readings.  Since then, we've used that same library to feed those idealized readings into our boards, then pull the computed orientation back into Matlab for analysis.  This is how we generated our datasheet parameters.

My trajectory simulation library was inspired by an academic project called IMUSim (you can google it).  At that time, IMUSim required a very specific (and not easy to reproduce), so I wrote this as a pure Matlab library.

So now my question to you and "the community", would this be a useful capability for Freescale to release?  If so, I'll have to pursue management approval, so it wouldn't happen immediately.   But knowing there is need for such a thing can help expedite things.

Regards,

Mike

[wingsof]

Michael,

are there any updates regarding trajectory simulation library? It`s definetely a highly demanded instrument for debugging and researching sensor data, because any other alternatives are missing or outdated.

Thank you!

[michaelestanley]

I admit, this one dropped off my radar.  Thanks for the reminder. I'll run it by my management.

Ran it by management and got approval to release.  I'll need to rerun unit tests and retrofit our open source license.  Expect a notice here when it is available for download (probably a week or so).

Regards,

Mike

[wingsof]

Any updates? :smileyhappy:

[michaelestanley]

Currently my own work queue is the only thing holding this up.  Before doing a general posting, I need to replace some 3rd party figures in the documentation with my own version, then go through legal review.  But if you are anxious to try it, send me a private note (mike.stanley@freescale.com).  I'll be happy to share a pre-publication version with you - especially if you'll give me feedback after trying it out. :smileyhappy:

[michaelestanley]

All,

Legal approved my updated user manual, and I have posted the full Matlab trajectory simulation library, documentation and examples to the MEMS Industry Group's GitHub site at memsindustrygroup/TSim · GitHub.

FYI, we used this library last summer to create trajectories which we used to characterize the sensor fusion library.  I expect to repeat the process soon on Version 5.00.  The general flow was to create trajectories, extract simulated sensor readings for those trajectories (which is completely automated), then pass those via serial port to a version of the fusion library modified to run from that data rather than real sensors.  The software ran on exactly the same MCU as production code, but with the added benefit of 100% known reference inputs and outputs.  Computed trajactories were then captured and passed back for analysis in Matlab.  The library here does not include the hardware interface.  But all trajectory functions are here and documented.

Enjoy!

Mike

[wingsof]

Wow, didn`t expect to get an answer so quickly!

Thank you, we will be waiting!

[jasonhaedt]

Mike,

I would be an advocate for such a feature as the more we can help to visualize the data and form contextual awareness about how the sensors are performing in our intended applications the better.  Let me know if you want to dicuss this more by emailing me at Jason.Haedt@avnet.com.

Thanks,

Jason Haedt