|
|
|
segmentation is not enough
Thursday, August 22 2019
Earlier in the week, I'd asked my boss Alex if I could maybe take Friday off to get an extra day this weekend at the cabin on Twenty Ninth Pond. He'd seemed okay with it if things went well this week. Unfortunately, though, this week happens to be one of the two or three such weeks every year that is crunch time in my line of work. Had Gretchen rented the cabin any other week, it probably would've been fine for me to play hooky on a Friday and start my Adirondack weekend early. But, alas, this was the week. And it wasn't long into today that I realized I would have to come in on Friday. Part of the problem was that Grumpy Jon, one of the few others on my team with deep institutional knowledge, was out sick and was apparently so sure that he would remain so that he'd blocked out the rest of the week on the shared Google Calendar. Complicating my week further was that this week I would have to be handling the biggest, most complicated data import of all the imports my Electron app does. The import was so big that it reliably crashed my app after working for an hour. The problem seemed to be a massive (greater than 100 megabyte) post of data from the frontend to the backend. It didn't take long to write a function to segment that data into chunks, and the segmenter worked on its very first use.
Unfortunately, my app still crashed (for some other reason) even with the segmenting in place. Since every build took an hour even to get the part where it crashed, you can imagine that debugging it was a slow, painful process. Especially so as I saw my Adirondack weekend dwindling down to nothing.
I ran a few more tests of the build back home after work as I experimented with the Waterbot. I'd given up on using the PCA9685 16-Channel Servo Driver as the source of PWM signals for the H-bridge driving the propulsion and steering motors and wired the H-bridge directly to the Raspberry Pi's GPIO pins. With some experimentation, I was able to write Python scripts to turn on the propulsion motor, turn off the propulsion motor, and then even turn the rudder left and right (though for some reason the rudder seemed to want to lean towards starboard). These scripts will require further tweaking, especially if I want to take advantage of the Raspberry Pi's PWM capabilities to make subtle changes in propulsive power or rudder orientation. But it won't take much more work to integrate these scripts with the cam viewing page that already exists, thereby allowing me to "drive" the boat remotely via WiFi. To make sure the boat was still seaworthy after loading it up with all the extra stuff, I ran some water into the tub and set the boat adrift. I don't think it will be able to handle rough waves, but it should be good enough for Twenty Ninth Pond, should I actually make it there this weekend.
Celeste is intrigued by the sea-worthiness test of the Waterbot in the upstairs bathtub.
For linking purposes this article's URL is:
http://asecular.com/blog.php?190822 feedback
previous | next |
