Outline of ICE-CEC2011 and Its Mechanism for Learning FPS Tactics
Seiji Murakami, Takumi Sato, Akihiro Kojima, Daichi Hirono, Naoki Kusumoto, Ruck Thawonmas · 2011
This paper describes an outline of ICE-CEC2011, our bot for CEC2011 humanlike bot competition. A salient feature of this bot is its mechanism for learning tactics. For this task, a system and GUI were developed that allow the player to control his or her character by only deciding which tactic to use under a given situation. Obtained log is used for training tactic selection of ICE-CEC2011 with Neuro Evolution of Augmenting Topologies (NEAT). Performance of ICE-CEC2011 is available in the contest site. 1. Acquisition of Tactic Log The player character in the log acquisition system has two states: item-collection and battle. In the former state, the character autonomously conducts its actions, as with the ICE-CEC2011 bot, according to pre-defined rules described in Sec. 3. In the latter state, however, its tactic is determined by the player; the rest is done automatically. In this work, we focus on four fighting tactics: 1) retreat (RT) by which the character retreats from the current opponent 2) approach (AP) by which the character charges the opponent 3) recover (RC) by which the character acquires the nearest curative item, and 4) hide-and-shoot (HS) by which the character hides behind wall and ambushes the opponent. The system allows the player to select one of these four tactics by clicking the corresponding GUI panel as shown in Fig. 1 or the corresponding key