Drones, Friends, and Navigation

Wednesday came and went and once evening hit I realized I had to switch things up. I had been letting myself worry about every small thing that needed to happen once I was on campus and how difficult everything would be once there… it was eating away at me. I’ve never felt homesick before, but this was as close as I would ever like to get again. I tried to remember back to my last trip here to figure out what was causing it and came to the conclusion that I wasn’t keeping myself nearly busy enough this time around. Two years ago it was wheels down at 8PM on a Sunday and work for a solid three months starting at 7AM the following Monday. So, I went into “recovery” mode and started scheduling things left, right, and center. I got to play “uncle” two times this week, the first with the niece of a now great friend, and the second with Daniel’s kids at the Drone Racing World Cup on Sunday. I can understand why someone would want to have kids if these types of interactions were the only ones they experienced. You don’t ever have to say no, the kids love you, and generally it is just a pretty great time!

And now… back to our regularly scheduled programming. Korea can be a beautiful and way less hectic place than you would ever think just visiting Seoul. I got to see scenic view after scenic view on my way down to the drone races on Sunday, mountains stretching up into the (what is probably pollution but let’s not worry too much about that) clouds in almost any direction you looked. Gone are the drab apartment buildings that stand to serve utility over all else and in its place a calm majestic landscape stretching as far as the eye can see. The day before this, I had the wonderful opportunity of visiting the Samsung Art Gallary which in a way almost prepared me for these new sights. Listening to the history behind each piece in the gallery, something I would normally find tedious, helped to place me in my new home. It also gave me a greater appreciation for the landscape of the trip I would take the following day.

This past week I also experienced something I realized I didn’t really miss at all. Packed public transportation. The issue with making public transportation so cheap and reliable is that everyone starts to use it, try to catch one of the last trains home and suddenly you find yourself with no room to move your arms. Now hold that mental image, add 34c (93.2f) weather to the mix and it starts to get a lot less pleasant. About the only part of your body that the A/C hits is your hair, and while I’m no biological expert, I’m pretty sure it doesn’t benefit too much from being cool. Luckily, there is a simple solution to this issue… don’t take the last train home. No matter if it is a Friday night or a Monday night it is guaranteed to be packed.

Speaking of navigation topics, I’m always pleasantly surprised what a conversation with my mentors from Yujin Robot ( the company I worked at last time I was in Korea) will turn up, and catching up this week did not let me down on this front. The main topic was SLAM (simultaneous localization and mapping), but more specifically, wouldn’t it be better to model navigation off of a more human approach? As a person, we don’t tend to think about navigation by saying “I’m 12cm away from the right wall and I’ve gone 12m down this hallway, so in another 2m I will need to turn 45 degrees and then move 20cm before turning another 8 degrees and arriving at the elevator.” What if instead of this method you didn’t worry about tracking precise locations but rather approached it much more like a human? You don’t care where you are in a hallway when you know at the end of it is the elevator you will take up to the next floor. Simply enter some sort of hallway navigation mode and once you are at the end of the hallway determine roughly where you are. While this is an extremely simple example that removes a lot of complexity and potential stumbling blocks, it does make for an interesting proposition. This approach would rely heavily on place recognition for localization which while not a solved problem has recently seen huge advances due to the advances in convolutional neural networks. The more difficult portion would be determining what mapping would look like in this type of setup. By removing the unnatural precision placement you also lose the ability to do a probabilistic/statistical approach to SLAM. Is a SLAM approach like this possible? I have no idea. But where is the fun in already knowing an answer!

comments powered by Disqus