Combat
Jump to navigation
Jump to search
Background Information
- What's special about it?
It has one of the best performances against rammers, thanks to anti-gravity movement.
Strategy
- How does it move?
- Multi-mode
- Wave Surfing/Wall Smoothing
- Old Skool anti-gravity movement boosted with precise prediction/wall avoidance
- How does it fire?
- Guess factor targeting with dynamic clustering segmented data.
- Using dynamic clustering/kernel density to estimate hit rate and choose bullet power. Working well in melee/team thanks to swarm targeting, shooting power 3 bullets against tight packs of bots, and owning rammers :) . Working acceptably against intermediate/weak bots in 1v1, but it is a waste of energy against surfers :( .
- How does it dodge bullets?
- Old Skool shrapnel dodging.
- Wave Surfing.
- Enemy virtual bullets generated with dynamic clustering segmented data in both modes.
- How does the melee strategy differ from one-on-one strategy?
- Anti-gravity in melee/team.
- Multi-mode in 1v1
- Wave Surfing at long range.
- Anti-gravity at close range (anti-ramming/dive protection).
- How does it select a target to attack/avoid in melee ?
- What does it save between rounds and matches?
- Between matches: Nothing.
- Between rounds: Wave data.
Additional Information
- Where did you get the name?
- From the Atari 2600 game Combat.
- Can I use your code?
- Not open-source.
- What's next for your robot?
- Cool algorithms
Energy management with dynamic clustering segmented data for hit rate estimation. Should be cool to watch.DONE- Improved precise prediction/wall avoidance. It is shaking (turning left and right) when moving near walls. Not cool to watch.
- Precise prediction/anti-gravity movement to calculate when to stop. It is vibrating (moving ahead and back) instead of stopping. Not cool to watch.
- Bomb sheltering evaluation to erase enemy virtual bullets.
- Rating improvement
Go to surfing for 1v1.DONEAnti-surfer gun for 1v1.DONECurve flattening for 1v1.DONE- Precise prediction/max escape angle calculation.
- Classification tuning.
- Fast math classes. Skipped turns are hurting the scores.
- A k-d tree to optimize dynamic clustering.
- What other robot(s) is it based on?
- Shadow (swarm targeting), DrussGT (energy management, DC classifiers), Diamond (gaussian kernel function).
- The code is 99% original. But I took most of the ideas from RoboWiki.
