Monday 1 July 2024

RoboCup 2024!


Ohne Holland Nach Holland Nach Noord-Brabant fahr'n wir zur WM!

    As the UEFA Euro Championship slowly but surely draws to it's climax, we're just getting started! On Sunday, July 14th, we're heading to Eindhoven for this year's RoboCup. The sting of last year's 0:9 final against B-Human was memorably unpleasant, at the German Open in April we managed a 4:1 against the northeners which isn't exactly a win, I know, but it didn't go unnoticed that we applied a considerable amount of pressure and ruined their perfect score balance, ha! In case you missed it:

 But yes, losing by a smaller margin is less bad but obviously not enough to dethrone the reigning world champion. So the question persists: What do? Let's highlight two innovations:

What's the T, Roboboy?

    Self-localization on the field is one of the principal problems of robofootball If the NAO doesn't know where it is, there is not much that can come from that. And if the robot communicates the wrong position, the break-down in efficacy just cascades through the entire team-strategy. Our new solution: finding T-shaped line sections on the field to boost the network processing the camera images. Sounds easy, right? No? Well, matter of taste. Oh and that means of course that you need to differentiate different kinds of T-lines, light conditions change all the time so color is strongly relative, the resolution is not exactly full HD and the images are tiny. So I guess it's actually quite tricky, nevermind. Did I mention that you need a substantial amount of training data? But the idea is very promising and our Machine Learning expert Tobias reports great results so far, so that's something to be excited about! We'll report how it goes!

Beep Boop, Baby!

    Another novelty is a true wild card - we've been working on a new communication protocol - using sound! This seemingly obvious choice for making yourself understood - humans have overall been relatively successful with that - comes however, with several caveats if you're talking robots. Whistles, referees, a rowdy crowd and a constant backdrop of chaotic noises from the robots themselves create a perfect storm of interference, so we had to get creative. The idea: if the interferences mainly consist of human-discriminable acoustic patterns - we need to pivot to physical properties that operate on an entirely different level: in chaotic interference, structured interference patterns are uniquely... unique. Spectacularly enough, quantitative musicology has quite a few things to say about that. Throw some symbolic communication strategies from theoretical linguistics into the mix, add a little dynamic data compression et voilá: Let's hope it works, fingers crossed! More updates to follow!


But wait, there's more!

As we get closer to the RoboCup, I'll post more about what we have been working on. 

Up next: This year's technical challenge Shared Autonomy: 2 against 2, one team fully autonomous, one team half autonomous, half human-operated. Will the humans and robots really work together? What could go wrong? Who will tell science-fiction writers that we're really starting to catch up?

No comments:

Post a Comment