Improving Generalization in Game Agents with Data Augmentation in Imitation Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yadgaroff, Derek, Sestini, Alessandro, Tollmar, Konrad, Ozcelikkale, Ayca, Gisslén, Linus
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911829302706176
author Yadgaroff, Derek
Sestini, Alessandro
Tollmar, Konrad
Ozcelikkale, Ayca
Gisslén, Linus
author_facet Yadgaroff, Derek
Sestini, Alessandro
Tollmar, Konrad
Ozcelikkale, Ayca
Gisslén, Linus
contents Imitation learning is an effective approach for training game-playing agents and, consequently, for efficient game production. However, generalization - the ability to perform well in related but unseen scenarios - is an essential requirement that remains an unsolved challenge for game AI. Generalization is difficult for imitation learning agents because it requires the algorithm to take meaningful actions outside of the training distribution. In this paper we propose a solution to this challenge. Inspired by the success of data augmentation in supervised learning, we augment the training data so the distribution of states and actions in the dataset better represents the real state-action distribution. This study evaluates methods for combining and applying data augmentations to observations, to improve generalization of imitation learning agents. It also provides a performance benchmark of these augmentations across several 3D environments. These results demonstrate that data augmentation is a promising framework for improving generalization in imitation learning agents.
format Preprint
id arxiv_https___arxiv_org_abs_2309_12815
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Improving Generalization in Game Agents with Data Augmentation in Imitation Learning
Yadgaroff, Derek
Sestini, Alessandro
Tollmar, Konrad
Ozcelikkale, Ayca
Gisslén, Linus
Machine Learning
Imitation learning is an effective approach for training game-playing agents and, consequently, for efficient game production. However, generalization - the ability to perform well in related but unseen scenarios - is an essential requirement that remains an unsolved challenge for game AI. Generalization is difficult for imitation learning agents because it requires the algorithm to take meaningful actions outside of the training distribution. In this paper we propose a solution to this challenge. Inspired by the success of data augmentation in supervised learning, we augment the training data so the distribution of states and actions in the dataset better represents the real state-action distribution. This study evaluates methods for combining and applying data augmentations to observations, to improve generalization of imitation learning agents. It also provides a performance benchmark of these augmentations across several 3D environments. These results demonstrate that data augmentation is a promising framework for improving generalization in imitation learning agents.
title Improving Generalization in Game Agents with Data Augmentation in Imitation Learning
topic Machine Learning
url https://arxiv.org/abs/2309.12815