14
Geoxion
149d

Most successful? Well, this one kinda is...

So I just started working at the company and my manager has a project for me. There are almost no requirements except:
- I want a wireless device that I can put in a box
- I want to be able to know where that device is with enough accuracy to be able to determine in which box the device was put in if multiple boxes were standing together

So, I had to make a real time localization system. RTLS.
A solo project.

Ok, first a lot of experiments. What will the localization technique be? Which radio are we going to use?
How will the communication be structured?

After about two months I had tested a lot, but hadn't found THE solution. So I convinced my manager to try out UWB radio with Time Difference Of Arrival as localization technique. This couldn't be thrown together quickly because it needed more setup.

Two months later I had a working proof of concept. It had a lot of problems because we needed to distribute a clock signal because the radio listeners needed to be sub-nanosecond synchronous to achieve the accuracy my manager wanted. That clock signal wasn't great we later found out.

The results were good enough to continue to work on a prototype.
This time all wired communication would be over ethernet and we'd use PTP to synchronize the time.

Lockdown started.

There was a lot of trouble with getting the radio chip to work on the prototype, ethernet was tricky and the PTP turned out to be not accurate enough. A lot of dev work went into getting everything right.

A year and 5 hardware revisions later I had something that worked pretty well!
All time synchronization was done hybridly on the anchors and server where the best path to the time master was dynamically found.

Everything was synchronized to the subnanosecond. In my bedroom where I had my test setup I achieved an accuracy of about 30cm in 3d. This was awesome!

It was time to order the actual prototype and start testing it for real in one of the factory halls.

The order was made for 40 anchors and an appointment was made for the installation in the hall.

Suddenly my manager is fired.

Oh...
Ehh... That sucks. Well, let's just continue.

The hardware arrives and I prepare everything. Everything is ready and I'm pretty nervous. I've put all my expertise in this project. This is gonna make my career at this company.

Two weeks before the installation was to take place, not even a month after my manager was fired, I hear that my project was shelved.

...
...

Fuck

"We're not prioritizing this project right now" they said.

...

It would've been so great! And they took it away.
Including my salary and hardware dev cost, this project so far has cost them over €120k and they just shelved it.

I was put on other projects and they did try to find me something that suited me.

But I felt so betrayed and the projects we're not to my liking, so after another 2-3 months I quit and went to my current job.

It would've so nice and they ruined it.
Everything was made with Rust. Tags, anchors, RTLS server, web server & web frontend.

So yeah, sorry for the rambling.

Comments
  • 2
    hmmm. couldn't you just have used RFID-triangulation with off-the-shelf components?
  • 3
    @tosensei with RFID you need to be close to the device already, within 1-2 meters I believe.

    UWB radio has a range of 50-150m so it would've been possible to track every device everywhere indoors at a ny time.

    You could look up the location from your PC and then walk to that location
  • 2
    That sounds like a quite interesting tech :)

    A bit like the apple airtags, andI can see that with the right marketing it could probably be quite popular in some industries.

    But the synchronization could be a problem I guess as I assume it needs calibration when setup?
  • 3
    @Voxera yes it does need calibration and the radio chip manufacturer describes the setup in which you can do that.

    However, I made it so it was auto calibrating.

    There are a lot of anchors and so every anchor is within receiving distance of a couple of other anchors.
    We know the position of every anchor.
    So when they synchronize to each other, we can calculate how far they think they are apart and compensate for that using the real distance they are apart.

    I don't remember the numbers exactly but I think it was roughly that an uncalibrated device had an accuracy of 10-20cm and calibrated around 1-2cm
  • 1
    @Geoxion to bad they did not at least try. I understand you disappointment
Add Comment