Wednesday Update
2 things. First, I've joined Lex Fridman on his 20k in a month challenge. The idea (was) to do the [number of likes under his post] pushups and pullups, but we have extended it to other exercises. My buddy Andrew will be joining, which is always nice since it will add up to the motivation factor.
Second thing, I'm really happy with the work I've been doing for the *mysteriously* Startup . My Machine Learning skills are really good, very happy with my performance, haha. *pats on the back*
Interesting Things
Don't want to say too much, but been talking to some interesting people about cool opportunities :P So in the last couple of days been working on some other projects to demonstrate my competence, which I did. The ML stuff I've been working on since December really paying off On another note, I kinda want to take another good online course. Udacity seems to offer a free month, so might try that.
Comments (0)
Wednesday Night
I wanted to gather some sensor data of a 'typical' longboard trip, to get a better idea of how much smoothening I would need to do. Check out the data for yourself: Typical Part of the Trip and Typical Full Trip Notice the 0 to 360 'jumps'. That's because polar angle is discontinuous. So right away we see that the raw data isn't that noisy (a couple of degrees), that's because I've purchased a sensor with a built-in Kalman Filter . In addition to it, I've also added my own tracking of an Exponential Moving Average which is frequently used in stock trading, but in a nutshell it's very similar to the Kalman Filter. So one observation right aways is that our EMA is lagging behind. Some lag is to be expected to get a better sense of a 'straight' direction, however, we can decrease the lag by tuning α parameter of the EMA.
Comments (0)
Tuesday Progress
I've decided that it would be best to go through everything from scratch on a new SD card. So far so good. Day: 1. BNO055 is working and displaying all of the 9 DOFs correctly (up to a constant) 2. Just installed and tested NeoPixels, so we are back to the shining. 2.5. Combined the two, seem to be working smoothly ~Unfortunately since I'm using a totally different sensor I'll have to spend some time writing a new interface class for it, but as I've found out last time it's totally worth it long term~ Evening: 3. Tested the compass + neopixel. The pixel closest to rotation angle lights up 4. Felt like still having a remote control would be nice to test different function 4.5. Implemented IR
Comments (0)