Playing FPS Games with Deep Reinforcement Learning

Authors: Guillaume Lample, Devendra Singh Chaplot

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show that the proposed architecture substantially outperforms built-in AI agents of the game as well as average humans in deathmatch scenarios.
Researcher Affiliation Academia Guillaume Lample, Devendra Singh Chaplot {glample,chaplot}@cs.cmu.edu School of Computer Science Carnegie Mellon University
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We use the Vi ZDoom platform (Kempka et al. 2016) to conduct all our experiments and evaluate our methods on the deathmatch scenario.
Dataset Splits No The paper specifies training and testing maps but does not mention a separate validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes All networks were trained using the RMSProp algorithm and minibatches of size 32. Network weights were updated every 4 steps, so experiences are sampled on average 8 times during the training (Van Hasselt, Guez, and Silver 2015). The replay memory contained the one million most recent frames. The discount factor was set to γ = 0.99. We used an ϵ-greedy policy during the training, where ϵ was linearly decreased from 1 to 0.1 over the first million steps, and then fixed to 0.1. We used a 16/9 resolution of 440x225 which we resized to 108x60.