ScalarR
- ScalarR Sub-pages:
- Version History
| ScalarR | |
| Author(s) | Xor |
| Extends | AdvancedRobot |
| Targeting | GuessFactor Targeting |
| Movement | Wave Surfing |
This page is under construction. For recent activities, see Version History.
Background Information
- What's special about it?
It's a melee bot, and shares same surfing algorithms in 1v1.
- How competitive is it?
No.1 in melee (as of 2021.6)
No.3 in 1v1 (as of 2021.6)
Strategy
- How does it move?
True Surfing in melee and 1v1, with different surfing paths generated. Danger computation considers Bullet Shadow, KNN and a simulation of simple guns (to encounter its weakness agains weak bots ;).
- How does it fire?
KNN/Play It Forward in melee, KNN/GuessFactor Targeting in 1v1. Energy Management is quite conservative that optimizes survival as main goal. The weights of 1v1 main KNN gun is tuned with Genetic Algorithms inspired by Skilgannon, by directly optimizing hit rate. The 1v1 anti-surfing gun is not tuned yet.
- What does it save between rounds and matches?
Everything between rounds, nothing between matches.
Additional Information
- Where did you get the name?
Scalar from Scalar Replacement, an optimization technique used in JVM to reduce GC pressure. Letter R is coined.
- What's next for your robot?
Maybe add a flattener ;) Which seems necessary against top bots.Done but need some anti-ram protection.- Tune gun / movement with more computation power. And maybe try some Gradient Descent and optimize predicted distribution instead similar to Kev ;)
- Looks like power management is the key to 100% PWIN, maybe some precise estimation or even RL works.
- What other robot(s) is it based on?
Directly and indirectly influenced by everything mentioned on this wiki.
- Experience learnt from ScalarN, ScalarBot and SimpleBot
- Understanding DrussGT. Thanks Skilgannon for sharing many details about how to build a competitive bot.
- Innovations since 2005
- Diamond’s code is very neat, I learned a lot but still get messy code.
- Understanding BeepBoop. Thanks Kev for sharing the idea about using gradients.
- to be listed...
