ESK8 WW Field Study. Statistical Analysis of Telemetry Data. METR / ACK / eSkate VESC etc

Hello fellow e-skate Builders,

This semester I will write my semester-thesis (similar to a longer paper that has to be presented before the Master thesis). I went to the Institute of Automotive Technology at my university (TUM, Munich) and approached them with a topic of my own that basically consists on doing an eskate telemetry analysis, but on a larger scale: aka field study. They found the idea very interesting. They also found eskates quite fascinating and loved the potential it has as an alternative means of transport, especially in larger cities.

Thanks to the immense potential of the VESC open-source architecture we can log tons of ride-information. That data can be later analyzed individually. My goal is to go one step further, which is statistically analyzing hundreds of logs simultaneously. Not focusing just on one rider but on our entire worldwide community.

Large automotive companies have been doing field studies for many years, focusing on the development of their cars. Some OEM’s such as Tesla for example constantly log every ride of each car worldwide.

The goal of this is to obtain a deeper knowledge of how we use our electric drivetrain. There aren’t many, or even any scientific studies that focus on estate drivetrains, and a big part of the forum knowledge is just based on personal “feeling” and experience.

Most of the forum builds are not comparable to prebuilt-estates, which in many cases (not counting Raptors) limit the way they are ridden, as they lack in power and range. Many of the builds are quite some beast witch rarely get to their power limits. This allows analyzing what really is required by experienced riders. At least in electrical terms :wink: .

What I have done till this point:

  • Preprocessing of the Data: I’ve written an e-Mail scraper (via google-script) for the study Gmail-address ([email protected]) that will automatically save all submitted ride data. It will also complete the database with the provided rider weight, as this is a critical factor on power consumption. All the raw data preprocessing steps are done on Matlab: Metr and Ack. log-files have to be structurally identical so that they can be simultaneously analyzed.

Future steps:

  • Processing of the data: The first step will be to plot several scattering diagrams to get a general view of the data. After that, I’ll prepare some more specific analyzes. I had planned to keep working with Matlab (as it was suggested by my tutors) but I recently spoke with someone who does data science for work and they suggested me to switch to Python. In his words: “using Matlab would be like trying to hammer a screw into a wall”

I wanted to integrate the telemetry data of the METR App and all the other apps that save their logs on a .csv format, such as the Ack. app and the eSkate-App for example.

Submitting the data (via Email) should be fairly easy: Mail: [email protected]

  • Metr App: Copy & Paste the record-URL (or URLs, the more the better) and write your ride weight (with gear) like this: weight=80kg // or weight=180lbs ((just don’t forget the “=” sign)) Alternatively connect iPhone to Itunes and share the .r ending Record-Logs. Also add the weight in the mail On android phones, it can be done via the file manager.

image macos: metr%20sharing android: android%20sharing

  • All the other Apps: Attach the .csv record files to the e-mail and add the rider weight like this: weight=80kg // or weight=180lbs ((just don’t forget the “=” sign)) image

General Information:

  • GPS Data will not be used (just the altitude values).
  • No email address will be stored. Once the data is saved all the emails are instantly deleted.
  • All will remain strictly anonymous.
  • Although It is a German University, the text will be written in English. This way everyone can read it.
  • The focus will lie on street builds (no emtb’s)
  • As for anything statistical the more datasamples are analyzed the better the results get at the end :smiley:

I would really appreciate suggestions (if you got any). If anything was left unclear please let me know.

Thanks for reading :slight_smile:

Best, Mathias


as suggested by akhlut I’ve added an optional G-Form for the board configuration. This allows to do more detailed research: (If you complete this form please remember to add your forum name followed by the board ID when submitting telemetry via e-Mail: @user1 for example)

Edit 2:

These files do not contain latitude/longitude values. So they are completely anonymus. More info:


@mods. can we sticky this?

also, given that you’re doing a statistical analysis consider anaconda which is python + R.


R was the other option so anaconda sounds great :slight_smile: . Thanks


i can see using python for data cleaning and transformations and R for the heavy lifting.


You might find this interesting. I’ll be submitting my logs. Can you take hundreds?

Oh I forgot to make a link! One sec

1 Like

oh yes, please send them :blush:

1 Like

Just wow! I wanna do something like this one day… I am still struggling with getting the most up to date parts for a remote(ervinelin remote). :frowning: A teacher I have at school that is also teacher at a FH I don’t know what a FH is called in English also suggested us to learn python as a basic language. I forgot the arguments why, sorry.


Good luck! If you are ready with everything, are you willing to share your analytics with us? :slight_smile:


of course, everything will be shared, even the final thesis


would you consider collecting board profile information as well? cell type, battery configuration, etc? if you provide us with a key that info can be stored in a separate file and we can use that key as a way to match our boards with the logs. it could be something as simple as our usernames here.


I wanted to keep is as simple as possible. People are lazy and don’t want to write. But, you are right. I could fish out the usernames (from those who want to share it) out of the email. And then prepare a G-Form to add board data.

Does it sound like a good Idea? Or better keeping it simple?


the more variable you have under control the smaller your error becomes. im assuming you’ll be running a series of multivariate linear regressions, and maybe some matched pairs depending on your sample size.


thank you @mmaner!


Pinned globally for a week, subject to @treenutter and @onloop approval.


Wow, I will do my best to keep it there. I hope this can be helpful for the entire community


Ah, I forgot to link what I intended with that post…

1 Like

Does METR and the other Apps take local weather in their data , that’s a very important factor that affects the data

no, I’m afraid that accessing the weather condition via the GPS data would be way to difficult

@akhlut does this look ok to you?