Shipometer

 Having given up on #SoME2, it's time to move on to the next project. Next week I will be going on a cruise. On the cruise, and also on the plane, I wish to record GPS signals. The pocketometer has all the right sensors for that, but unfortunately is not sufficiently reliable. The Raspberry Pi has already proven itself capable of recording GPS from one of the ZED-F9R breakout boards. Now the question is if it can record the sensors on I2C, and the time pulse.

I want to see if I can do this with just the parts that I already have on my desk.

I have a belt bag big enough to hold all the sensors, the Pi, and a 20Ah USB battery pack. That would be a lot less suspicious than stuffing stuff in my pocket.

It would also be great if the Pi could act like a wifi hotspot and serve SSH through it. That way I could look at it on the phone while (literally) in flight.

The last thing that would be awesome is timer capture on the GPIO, of at least the PPS and maybe others, like the interrupt lines from the sensors. If it can't, maybe we could get a program on the Teensy that would do the timer capture and output on UART or as an I2C slave.

#SoME2 post-mortem

 I did not get a video out on time for SoME2. Even for just the descoped "good part" video, I couldn't get it done in time this morning.

Deadline pressure

I ended up going with the Kalman gain video. I will use this blog as "making of" documentation.

• I am going to use PictureBox much more than Manim -- I have already forgotten how to use Manim. I think I will only need it for dancing equations, and I don't plan on using those much. MatPlotLib knows how to use TeX, so it can make nice-looking math, but can't make them dance as well.
• One video, two videos, N videos? I have about 9 minutes of narration so far, and have just covered up to measurement space. It might take 20 minutes to cover the stuff I want just for linear Kalman filtering. KalmanGain implies that the most important or interesting stuff is calculating the gain. That by itself might only take a minute, but all of the pre-requisites might take even more than 20 minutes. For now, I am planning on one long video, covering transformation of uncertainty from state to measurement space and back, and the hand-waving justification for why there even is an optimal gain matrix to transform innovations back into state vector adjustments.
• Pure video, or text plus video? This interacts with above. The hard thing about videos, is that the maker of the video always has to leave stuff on the cutting room floor. This leads to a contradiction: If I include everything I know, the video will be too long and boring, while if I don't, the pedants in the audience will use this as an opportunity to show how much smarter they are than me. Text plus video would allow me to put the most important bits in a video, which could be run as a continuous playlist. The text part would then include the videos, and footnotes with all the pedantry included.
• Narration. I hate the sound of my recorded voice. Plus, I don't have a good audio setup yet. Therefore, I am using Amazon Polly to generate narration from a script. I am having neural British Amy read it. This is the first text-to-speech I have heard that I would say is good enough to be plausible as a human-read narration. I would say it's 80% of the way there. I just wish I could adjust the emphasis on some words. American Joanna is also good enough, but as an American, I at least subconsciously buy into the "intelligent British accent" stereotype. I know that I write slightly different for Amy than I would for myself, and a lot different than I do on this blog.

Which brings us to deadline pressure. SoME2 was announced in the middle of June, but I didn't start on it in earnest until this past Monday (August 11). If I had all the time in the world, I would make a video covering time updating, nonlinear models, and maybe various filters beyond Extended Kalman. It might take an hour, but I would break it up into roughly 20-minute chapters. With the deadline pressure, I am not going to be able to carry out this whole program. I might be limited to just the gain matrix.

I am *not* going to do things like publish one chapter before the #SoME2 deadline and extend a playlist afterwards. I am committing to publishing a complete thought, or nothing at all.

Why is #SoME2 even important? For one thing, I don't expect to win. This might be a winning concept in the right hands, but I am learning that my hands are not those. I am also learning why 3b1b only publishes as infrequently as he does. That's the non-reason, so the reason is: I like having large numbers in my YouTube channel. My last-years late #SoME1 got 104,000 views, even though it didn't make the official #SoME1 playlist.

Also, without deadline pressure, I may never actually make this. One of the issues with my projects is that whenever I am working on one, I wish I was working on another. I'm sure that's a common character flaw among humans in general, but I don't know what it's called or how to fight it.