Deep Reinforcement Learning for Playing 2.5D Fighting Games

TL;DR

This paper introduces a novel deep reinforcement learning approach using an A3C+ network with recurrent layers to effectively learn and play 2.5D fighting games like Little Fighter 2, overcoming visual ambiguity and complex action sequences.

Contribution

It presents a new A3C+ network architecture with recurrent info layers tailored for 2.5D fighting games, and develops an OpenAI-gym-like environment for training and evaluation.

Findings

The model successfully learns to play Little Fighter 2 in various settings.

Recurrent info layers improve recognition of combo skills and game states.

The approach demonstrates effective handling of visual ambiguity and sequential actions.

Abstract

Deep reinforcement learning has shown its success in game playing. However, 2.5D fighting games would be a challenging task to handle due to ambiguity in visual appearances like height or depth of the characters. Moreover, actions in such games typically involve particular sequential action orders, which also makes the network design very difficult. Based on the network of Asynchronous Advantage Actor-Critic (A3C), we create an OpenAI-gym-like gaming environment with the game of Little Fighter 2 (LF2), and present a novel A3C+ network for learning RL agents. The introduced model includes a Recurrent Info network, which utilizes game-related info features with recurrent layers to observe combo skills for fighting. In the experiments, we consider LF2 in different settings, which successfully demonstrates the use of our proposed model for learning 2.5D fighting games.