# Learning to Simulate Dynamic Environments with GameGAN > GameGAN learns an action-conditioned neural simulator of a game from recorded frames and the actions taken, without access to the game’s logic or engine. An action-conditioned LSTM dynamics engine with a stochastic latent variable predicts the next state, an optional external memory module builds an internal map so that revisited locations stay consistent, and a rendering engine decodes frames while separating static from dynamic components. Training is end-to-end with adversarial losses, including an action-conditioned discriminator, and a temporal cycle loss for memory consistency. The authors evaluate on a modified Pac-Man, a ghost-free Pac-Man-Maze and VizDoom “take cover”, training RL agents inside the learned simulators and testing them in the real games, and measure long-term consistency with a come-back-home task; a human-playable model of the official Pac-Man is shown qualitatively. - Authors: Seung Wook Kim; Yuhao Zhou; Jonah Philion; Antonio Torralba; Sanja Fidler - Venue: CVPR 2020 - Status: Conference paper - Publication / posting date: June 2020 - Date source: IEEE’s Crossref deposit and an archived copy of the IEEE Xplore page give June 2020, represented at month precision. The May 2020 arXiv submission and the conference dates are not substituted. - First arXiv submission: May 25, 2020 - arXiv: 2005.12126 - Reviewed arXiv version: v1 - DOI: 10.1109/CVPR42600.2020.00131 - Sources reviewed: 2026-10-19 - Added to the catalog: 2026-10-31 - Original paper: - PDF: - Code: - BibTeX: - Catalog page: - Record JSON: - Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-gamegan/index.md) ## Scope and limitations GameGAN imitates a game’s pixels and dynamics; it does not recover game code, symbolic rules, assets or an executable. Training needs large logged datasets with actions: 40K Pac-Man episodes from a high-entropy DQN agent that sees the full environment, and 10K random-policy VizDoom episodes, at 84×84 and 64×64 resolution. Quantitative results cover small grid-based Pac-Man variants and one VizDoom scenario, with an external classifier supplying game outcomes. Agents trained in the simulators can exploit modeling errors: the full model sometimes let Pac-Man pass through walls, and its Pac-Man agents scored below those of the memoryless variant, while on VizDoom the World Model baseline scored higher than GameGAN. The released code supports VizDoom training only, relies on a deprecated WorldModels data setup, and is under the non-commercial NVIDIA Source Code License; Pac-Man data and the official Pac-Man model are not included. ## Why it fits this collection Adds a neural approach to reconstructing a playable game simulator from observed frames and actions, complementing the section’s program-synthesis and black-box game-replication research. ## Publication and review notes Crossref and the archived IEEE Xplore page give the five authors, CVPR 2020 and pages 1228–1237; IEEE metadata capitalizes the title as “Learning to Simulate Dynamic Environments With GameGAN”, while the CVF open-access paper and arXiv print “with”. The 10-page CVF PDF is the Computer Vision Foundation’s open-access version; IEEE Xplore holds the version of record. arXiv has only v1, submitted 2020-05-25 with the comment “CVPR 2020”. The repository README at the reviewed 2021-11-11 commit gives BibTeX pages 1231–1240, which differ from IEEE, and its Project Page link points to an unrelated lift-splat-shoot page. The README lists Ubuntu 18.04 and Python 3.8.3, and requirements pin torch 1.4.0 and torchvision 0.5.0. LICENSE section 3.3 limits use to non-commercial research or evaluation. ## Primary sources 1. 2. 3. 4. 5. 6. 7. 8. ## About this entry Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX. Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.