Subscribe via Email

Wednesday, October 12, 2016

IROS 2016: Robot Magic! Part 1 - Welcome to Korea

Post #24 of "Today In The AA" series. (~8 minute read)   Click to go to Part 2 of this post!

Arrival In Korea

I arrived in Seoul on Saturday evening and took a 3 hour bus to Daejeon from the Incheon International Airport.

A bed of Skyscapers tile the neo-Korean city landscape, giving a futuristic sci-fi ambience as we approach from the perimeter. Upon arrival I was greeted by many rounded symbols, none of which were English. Along with these symbols I was greeted by a friendly taximan who gestured me to board. I was indirectly given a tour of Daejeon as we navigated the various interlinked streets searching for my hotel. Roughly 20 minutes later after stopping at Toyoko Inn, one of the welcome receptionists called the hotel I'm actually supposed to be at and gave the taxi man directions. Yeah, you're reading this right. I'm standing in a Japanese hotel lobby with my cab driver and a panel of Korean receptionists, waving my hands around with an airport sleep pillow wrapped around my neck. Thankfully exhaustion is a universal language, and they figured I wasn't about to go hiking up a mountain (you'd be surprised...).
I got back to the taxi and navigated to the hotel. Out of all this assistance and time driving, he charged me 7500 Won, which is roughly $8.74 Canadian. I was expecting a couple more zeros in that number, but instead I was assisted out of the cab with nothing but a friendly smile from the driver and my phone I almost forgot in the cab.

I ensured didn't drop any of my belongings on the dim lit road of dunsan-ro. I entered my room, plucked in the key card I was given, and the room came to life.


Goodstay Samjoha Motel suite
Goodstay Samjoha Motel suite
It kind of has that romantic vibe you see in corny films where the dude flicks a switch and the room transforms into a shine of romance. This room was more designed for utility, but I mean hey if you have a minifridge and a PC in your room that's romantic enough for me.
I capitalized on sleep the next day and caught up on studies (since I was just coming from another conference). Let's fast forward to Monday, where the magic begins.


IROS Day 1 (Preparation)


Streets of Daejeon South Korea walking with Robots blog.
Streets of Daejeon just outside of hotel.
I hopped a cab to the Daejeon Convention Center and arrived by 7:15am. The venue is beautiful, again with Smart City skyscrapers lining the streets. Setup was taking place for the conference, so I navigated my way to Air Dome C, where the competitions would be held.
IROS 2016 Conference events
Outline of events at IROS
IROS 2016 Entrance to Daejeon Convention Center
Entry way to Daejeon Convention Center
IROS 2016 Welcome banner
Welcome banner!
Magic show setup IROS 2016
Magic show setup 
I entered the dome and was alone, unsure as to how early I was. A table was setup with a projector and I deduced this was for the competition. I took the keener approach (as usual) and began setting up my equipment right in front of the projector. Jacky was the next to arrive by 9am, and let me know the table I set up at was the judging table. I promptly moved my equipment and setup at what is now Snobots HQ.


The other teams starting streaming in, first the Taipei Snipers and then Qiron Robotics.

Food!
We all began setting up tricks and quietly practising. It's exciting yet terrifying to see your opponents rehearsing next to you. It's like being on death row, knowing you have 1 day left to live, and watching the chefs prep your last supper of choice. Well... I've never been on death row, at least not that I remember. But me and other researchers, at least we get along. We go well together, we goes together like peas and carrots. I'm sorry I really stretched for that reference. I'm just hungry right now... speaking of which breakfast is served upstairs each day! I'm usually the guy who hounds the food table as if it's going out of style.

Most of my day was spent focusing on playing card recognition for my trick. I needed the Darwin OP2 I was using to correctly classify cards from a standard deck. I worked several hours of the day adding new configuration settings for dilation, contouring, and other ways to handle the changing brightness in the room.

Practising at home was simple enough; but on competition day out here I'll be required to bring the robot to the front of the audience, and have it correctly classify cards while there is a glaring light from the projector light on it, and the sun being all bright and sunshiny.

Alongside the lighting issue, there was practice for a drone manoeuvring competition going on in the area immediately next to us.


These things are loud as you can probably imagine. Much like loons with air-horns for beaks, they make a loud droning sound as they hover through obstacles before hitting an object and screeching towards the ground to explode into several pieces. I don't think that part is intentional; but it's so amusing to watch. This does pose as a difficult speech recognition challenge though... My robot is required to hear me talk to it, and even with a good quality microphone there is severe obstruction with the drones flying about. This added challenge makes the whole competition more interesting in my point of view. By now you can also deduce which teams are using vision and speech... we're all lining the drone competition stadium... watching -- straight faced -- impressed, yet desperately wanting to give their team leader an uppercut.

Come evening, there was a snazzy conference dinner just outside of the convention centre.
Water show! Or perhaps another drone crashed into the pond?
I camped out with the Taipei Snipers and we enjoyed some fantastic Korean cuisine. I didn't realize until now how popular spicy food is here. That's a big difference from Taiwan, where most students I talk with refrain from spicy goods.


We took a nice walk back to the hotels together and I worked on vision throughout the night, tuning the robot with 1 day left to practice...

IROS Day 2: Last day for preparation

I went for a nice walk to the convention centre enjoying more of the scenery now that I actually know where I'm going. Upon arrival by around 9am I resumed vision and speech tuning.


Beautifu scenery walking to Daejeon convention center South Korea
Beautiful scenery walking to convention center.
I had to start thinking in a more clever way now. With 1 day remaining, I had no time to implement some elaborate system to adjust for dynamic lighting. I was thinking of ways to ensure that if I randomly pick 3 cards from the deck, my robot will correctly classify them... Upon checking through cards, it seemed that usually Spades were failing (being classified as diamonds...) so one option was to just remove spades from the deck. This conflicts with my research report in which states I used each card in the deck for my trick. I need to find a way to improve the accuracy without removing cards. Along with this, if I slightly adjust lighting, such as by going up on the stage, suddenly Spades are recognized, and now Clubs are misclassified as Spades... This is the result of using gray-scaled pixel intensity [0-255] values for classification... Much like the MNIST data set, it works well if your input is standard; but under changing brightness you get skewed results.

Team HURO Arrives later in the morning and began setting up what looks to be quite an elaborate trick. They refrained from performing live in front of us. I figure they either have a really rock solid presentation ready, or have nothing ready.

Later in the evening the finalist competition times were posted:

Qiron Robotics: 8:30am
Seed Robotics: 9am
Snobots: 9:30am
Taipei Snipers: 10am
HURO: 10:30am

I tried to head back earlier to catch up on some rest, but was mainly focused on vision tuning yet again, trying to decide on the best approach to ensure classification worked.... No matter what... on live competition day, the 3 cards need to be classified correctly, or else it would fail in front of the whole audience, and butcher my chances of placing well. Let's see how things turn out...

Part 2 >> 

No comments:

Post a Comment

Please feel free to give feedback/criticism, or other suggestions! (be gentle)

Subscribe to Updates via Email