DeBroglie/Archived Talk 2010

Shooting where a surfer would go

This is not as simple as it first seems... Sure, you can keep track of surf stats from the enemy perspective, and when you fire, you can shoot at a less visited bin instead of a most visited one. But most of the time, there are already 1-3 bullets in the air. The enemy will first surf each of those before deciding how to surf the bullet you're firing, then he'll choose the safest from the bins that are still reachable, not from all of them. Add in the discrepancies in data analysis (what you segment on, how fast you roll your stats) and it's very hard. I don't think anyone's been able to find something that works. I've personally tried quite a few approaches with no success. Good luck though. =) It's actually something that I still think has potential. --Voidious 13:06, 27 October 2010 (UTC)

• That is something that occurred to me in the conceptual phase. My plan for movement was to recursively surf all waves in the air (or up to whatever depth I can handle, I have little experience with where Robocode shuts a bot off on data processing for a turn) meaning that by symmetry I could fire at an opponent surfing all of the waves I had in the air. In reality, this would work delightfully if I pitted two copied of deBroglie against each other, but the differences between my surfing algorithm and any other bot's would make the exercise fruitless in practice. -Tkiesel 17:49, 27 October 2010 (UTC)
• As a one-off, I wanted to try stepping the target through all the waves I have in the air assuming the target will pick the safest direction for each. I thought this may give a much more limited range of possible angles to fire at. These are all ideas that have been done to death by others, I'm sure, but I'm enjoying the process of trial, error and discovery. - Tkiesel 17:50, 27 October 2010 (UTC)
• Ah, I see. Well, if you could simulate enemy surfing, I personally think it could perform well, despite the differences in surfing styles and stats. The problem there is just CPU time. In normal "True Surfing", each tick you just make one decision (forward/backward/stop). But to aim, you'd have to simulate that decision over a lot of ticks, which is an order of magnitude more calculations.
  I like your idea about assuming the safest direction at each wave. I think you need something like that - simpler than full surfing simulation but taking stats into account - to make this work. My best idea in this department is near the top of Archived talk:Dookious 20071111 if you want to take a look. Hope you have more success than I did. Good luck. :-) --Voidious 18:08, 27 October 2010 (UTC)
  • I think I have an alright grasp of the most basic form of wave surfing. But in case I don't, I'll ask: I know a simple wave surfer will make the forward/backward/stop decision on every tick, but effectively, this means the surfer will either head to the far end of its possible bearing range, or stop/oscillate around a bearing on the wave that balances the forward/backward danger. If the surfer uses nothing for movement but the closest wave, then it should be trivial for that surfer to calculate this destination ahead of time in one tick (goTo style) and produce the same motion, yes?
    I like the masking concept you applied to this problem in the talk archive you linked! Once I have this bot halfway competitive, I'll have to look into exploring these ideas! For now, I really have to figure out what statistics are best to segment on, and get my bot wave surfing! I get hit wayyy too easily right now. -Tkiesel 21:05, 27 October 2010 (UTC)