Text-based adventures of the golovin AI agent

Kostka, Bartosz; Kwiecieli, Jaroslaw; Kowalski, Jakub; Rychlikowski, Paweł

doi:10.1109/cig.2017.8080433

Cited by 27 publications

(36 citation statements)

References 22 publications

(29 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We evaluated BYUAgent 2016, Golovin, CARL (BYUAgent 2017), and NAIL on each of the 20 test games for 1000 game steps. This is to make our results comparable to those in [18], [20] with the same number of steps. The results reported here are averages over 10 such runs for Golovin and NAIL; unfortunately, we were only able to do one run for each of the BYU agents (discussion below).…”

Section: Byuagent 2016)mentioning

confidence: 59%

“…This is particularly impressive considering that NAIL and CARL are arguably using less domain-specific knowledge than Golovin, e.g. Golovin's "battle mode" [20]. It is possible that this domain-specific knowledge overfits to the training set used by the authors, and generalizes less well to the games in our test set.…”

Section: Resultsmentioning

confidence: 92%

“…The Golovin 4 agent [20] uses command generators to propose a non-empty set of commands for a given game environment by inserting nouns taken from the game's narrative text into "command patterns" -a set of 250,000 verb phrases extracted from various game walkthroughs, tutorials, and raw narrative text. To do this, a word2vec model trained on 3000 fantasy books is used to propose synonyms for each noun using n-best cosine similarity.…”

Section: B Golovin (2017)mentioning

confidence: 99%

“…The games were written in different styles by different authors, and some had been previously either commercially released, or submitted to interactive fiction writing competitions. There is no overlap between our test set and the training set used by [18], [20]. Unlike some of the games in this training set, all games in our test set provide a numerical score that allows for a fine-grained measurement of game playing performance.…”

Section: Byuagent 2016)mentioning

confidence: 99%

See 3 more Smart Citations

The Text-Based Adventure AI Competition

et al. 2019

View full text Add to dashboard Cite

In 2016, 2017, and 2018 at the IEEE Conference on Computational Intelligence in Games, the authors of this paper ran a competition for agents that can play classic text-based adventure games. This competition fills a gap in existing game AI competitions that have typically focussed on traditional card/board games or modern video games with graphical interfaces. By providing a platform for evaluating agents in textbased adventures, the competition provides a novel benchmark for game AI with unique challenges for natural language understanding and generation. This paper summarises the three competitions ran in 2016, 2017, and 2018 (including details of open source implementations of both the competition framework and our competitors) and presents the results of an improved evaluation of these competitors across 20 games.

show abstract

Section: Byuagent 2016)mentioning

confidence: 59%

Section: Resultsmentioning

confidence: 92%

Section: B Golovin (2017)mentioning

confidence: 99%

Section: Byuagent 2016)mentioning

confidence: 99%

See 2 more Smart Citations

The Text-Based Adventure AI Competition

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Several works [2], [4], [7]- [9], [11], [18] also build agents for text-based games based on the DQN approach designed for action video games [1]. One key consideration when learning to play text-based games is how to represent game states.…”

Section: Related Workmentioning

confidence: 99%

Comprehensible Context-driven Text Game Playing

Yin

May

2019

2019 IEEE Conference on Games (CoG)

View full text Add to dashboard Cite

In order to train a computer agent to play a textbased computer game, we must represent each hidden state of the game. A Long Short-Term Memory (LSTM) model running over observed texts is a common choice for state construction. However, a normal Deep Q-learning Network (DQN) for such an agent requires millions of steps of training or more to converge. As such, an LSTM-based DQN can take tens of days to finish the training process. Though we can use a Convolutional Neural Network (CNN) as a text-encoder to construct states much faster than the LSTM, doing so without an understanding of the syntactic context of the words being analyzed can slow convergence. In this paper, we use a fast CNN to encode positionand syntax-oriented structures extracted from observed texts as states. We additionally augment the reward signal in a universal and practical manner. Together, we show that our improvements can not only speed up the process by one order of magnitude but also learn a superior agent.

show abstract

TextWorld: A Learning Environment for Text-Based Games

Côté

Kádár

Yuan

et al. 2019

Communications in Computer and Information Science

105

View full text Add to dashboard Cite

We introduce TextWorld, a sandbox learning environment for the training and evaluation of RL agents on text-based games. TextWorld is a Python library that handles interactive playthrough of text games, as well as backend functions like state tracking and reward assignment. It comes with a curated list of games whose features and challenges we have analyzed. More significantly, it enables users to handcraft or automatically generate new games. Its generative mechanisms give precise control over the difficulty, scope, and language of constructed games, and can be used to relax challenges inherent to commercial text games like partial observability and sparse rewards. By generating sets of varied but similar games, TextWorld can also be used to study generalization and transfer learning. We cast text-based games in the Reinforcement Learning formalism, use our framework to develop a set of benchmark games, and evaluate several baseline agents on this set and the curated list.

show abstract

Text-based adventures of the golovin AI agent

Cited by 27 publications

References 22 publications

The Text-Based Adventure AI Competition

The Text-Based Adventure AI Competition

Comprehensible Context-driven Text Game Playing

TextWorld: A Learning Environment for Text-Based Games

Contact Info

Product

Resources

About