Difference between revisions of "Wraith/Version History"

From Robowiki
Jump to navigation Jump to search
(Restored edits authored by Wolfman dated 2019-12-07T21:07:56+00:00)
 
(No difference)

Latest revision as of 01:28, 25 April 2020

Wraith Sub-pages:
WraithVersion History - Challenge Results - Wolfman's TODO List


V 0.3

RoboRumble: -??

  • Removed all old guns from the gun array firing at opponents in favour of a single DC gun while it gets tuned to the max.
  • Much improved calculation of Maximum Escape Angle that takes into account: Opponent velocity, heading, walls, bot width. Utilises non-smoothed precise prediction.
  • New DC gun is not yet optimised, very slow initialisation (skipping turns on first init) and is slow after that - but will be improved with upcoming work.


V 0.20

RoboRumble: - 177th APS ~69.27

  • Added GuessFactor gun to virtual gun array for firing at opponents.
  • Other small tweaks.


V 0.17

RoboRumble: - 283rd APS ~64.39

  • Remove simple circular gun from VGA.
  • Fix bug in detection of power 3.0 bullets fired by opponent.
  • Massive optimisation (~150% faster).
  • Increased DangerPrediction iteration count to 3 per frame from 2.
  • New stationary movement mode when opponent is dead or disabled.
  • New colour scheme to differentiate from AgentSmith and AgentSmithRedux

Author's notes: Turns out that it was not the simple circular gun reducing the APS by 2%, it was only doing it in my set of test bots!.

V 0.15

RoboRumble: - 281st APS ~64.47

  • Fix crash bugs
  • Fix occasional crash into walls
  • Optimisations to make it run faster
  • Added simple circular gun to the gun array
  • Added future bullet prediction to the DangerPrediction. Currently predicts +1 bullet ahead based on target gun heat.

Author's notes: Turns out that adding the simple circular gun to the VGA actually decreased the score by 2%.

V 0.11

RoboRumble: - 224th, APS ~66.61

  • Fix Robocode version to match RoboRumble.

V 0.1

RoboRumble: - ??? (Wrong robocode version)

  • Very early release
  • Refactor of AgentSmithRedux internals for handling of history & prediction positions to use absolute ticks instead of relative. Makes reasoning about mismatches in tick values easier to understand and track down bugs.
  • On rumble to ensure no major bugs introduced in refactor so should be similar score to AgentSmithRedux if it goes well!