|Targeting||Dynamic Clustering, GuessFactors|
- What's special about it?
- It's a perceptual duelist with a gun evolved by genetic algorithms. And it doesn't "cheat" by leveraging state-tracking things like
getDistanceRemaining(), like most old school perceptual bots do. (That was allowed under the rules of David Mccoy's perceptual survivalist league, but it seems to me quite contrary to the spirit of a bot that doesn't store any state!)
- How competitive is it?
- Quite competitive - the only perceptual bots it loses to are NanoVolver (badly) and PerceptualDuelist (sometimes). With the rock/paper/scissors nature inherent to such a limited format, I'm pretty happy with that for now.
- How does it fire?
- This is the interesting part of the bot. It has a data space (encoded in a ~4000 character string) using a few attributes that don't require tracking state change. It looks up the nearest point to the current situation and fires at the GuessFactor stored for that point. The locations and GuessFactor values for these data points were evolved using genetic algorithms with WaveSim against most of the RoboRumble population.
- It could be the strongest non-learning gun ever. =) (LittleBlackBook?) It definitely outperforms Head-On Targeting, Linear Targeting, and Circular Targeting (which isn't perceptual anyway).
- What does it save between rounds and matches?
- Nothing - it doesn't even save anything between ticks!
- Where did you get the name?
- RetroGirl is one of the main superheroes from Powers, my favorite comic book series. And it seemed kind of appropriate since nobody's making perceptual bots any more.
- Can I use your code?
- Yes, it's released under the zlib license.
- What's next for your robot?
- I'm not sure. Beating all other known perceptual bots would be very satisfying, but I may focus on general RoboRumble performance. Evolving the movement with GA would also be a cool endeavor.
- Does it have any White Whales?
- NanoVolver and PerceptualDuelist.