awesome-world-models-for-digital-agents

KO | EN

gitlite — search

#python #machine-learning #android #tensorflow #docker #golang #deep-learning #python3 #computer-vision #mongodb #hacktoberfest #angular

awesome-world-models-for-digital-agents

★ 50 Open GitHub ↗

Digital Agents Meet World Models: A Survey

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Awesome World Models for Digital Agents

[![Awesome](https://awesome.re/badge.svg)](https://awesome.re) [![Paper](https://img.shields.io/badge/Paper-PDF-red)](https://raw.githubusercontent.com/Darwin-Agent/awesome-world-models-for-digital-agents/main/paper.pdf) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://raw.githubusercontent.com/Darwin-Agent/awesome-world-models-for-digital-agents/main/LICENSE)

A curated list of world models for digital agents across games, web & GUI, tool use, and code. Our [survey paper](paper.pdf) introduces a unified design space **W = (X, L, U)** — what is modeled (**X**), how it is built (**L**), and how it is used by the agent (**U**).

Papers are listed in reverse chronological order within each section.

## 📢 News & Updates

- **2026-05** — Survey paper is out! Check out [**Digital Agents Meet World Models: A Survey**](paper.pdf). arXiv link coming soon!
- **2026-04** — Repo is live! 100+ papers on world models for digital agents, organized by domain.

## Table of Contents

- [🗂️ Taxonomy Overview](#%EF%B8%8F-taxonomy-overview)
- [🎮 Games](#-games)
- [🌐 Web \& GUI](#-web--gui)
- [🔧 Tool Use](#-tool-use)
- [💻 Code](#-code)
- [🏋️ Benchmarks \& Environments](#%EF%B8%8F-benchmarks--environments)
- [📊 Datasets](#-datasets)
- [📚 Related Surveys](#-related-surveys)
- [🤝 Contributing](#-contributing)
- [📖 Citation](#-citation)

## 🗂️ Taxonomy Overview

The survey organizes world models for digital agents along a unified design space **W = (X, L, U)**:

| Dimension | Question | Key Choices |
|:----------|:---------|:------------|
| **X: Model Specification** | *What is modeled?* | State deltas, Full observations, Latent transitions, Auxiliary outcomes |
| **L: Learning & System Realization** | *How is it built?* | LLM-based, Visual generative, Latent dynamics, Code-based, Hybrid/modular |
| **U: Agent-Facing Role** | *How is it used?* | Decision-time planning, Training-time interaction, Predictive verification |

## 🎮 Games

World models for game environments, including classical model-based RL, visual game generation, and text-interactive worlds.

### Latent Dynamics & Model-Based RL

+ [**DreamerV3**](https://arxiv.org/abs/2301.04104) (Nature, 2025) — RSSM + symlog loss; generalizes across 150+ diverse tasks. [[GitHub]](https://github.com/danijar/dreamerv3)
+ [**DIAMOND**](https://arxiv.org/abs/2405.12399) (NeurIPS, 2024) — U-Net diffusion transition operator for Atari. [[GitHub]](https://github.com/eloialonso/diamond)
+ [**STORM**](https://arxiv.org/abs/2310.09615) (NeurIPS, 2023) — Stochastic Transformer + VAE world model. [[GitHub]](https://github.com/weipu-zhang/STORM)
+ [**IRIS**](https://arxiv.org/abs/2209.00588) (ICLR, 2023) — VQ-VAE + Transformer autoregressive world model. [[GitHub]](https://github.com/eloialonso/iris)
+ [**EfficientZero**](https://arxiv.org/abs/2111.00210) (NeurIPS, 2021) — MuZero + self-supervised consistency; sample-efficient. [[GitHub]](https://github.com/YeWR/EfficientZero)
+ [**Dreamer**](https://arxiv.org/abs/1912.01603) (ICLR, 2020) — Latent imagination via RSSM; multi-step backpropagation. [[GitHub]](https://github.com/danijar/dreamer)
+ [**MuZero**](https://arxiv.org/abs/1911.08265) (Nature, 2020) — Value-aligned latent dynamics with MCTS; masters Go, chess, shogi, and Atari.
+ [**PlaNet (RSSM)**](https://arxiv.org/abs/1811.04551) (ICML, 2019) — Learning latent dynamics for planning from pixels. [[GitHub]](https://github.com/google-research/planet)
+ [**World Models**](https://arxiv.org/abs/1803.10122) (NeurIPS, 2018) — VAE + MDN-RNN; influential early architecture. [[GitHub]](https://github.com/hardmaru/WorldModelsExperiments)

### Visual Game Generation

+ [**GameNGen**](https://arxiv.org/abs/2408.14837) (ICLR, 2025) — U-Net diffusion runs DOOM at 20 FPS.
+ [**The Matrix**](https://arxiv.org/abs/2412.03568) (arXiv, 2024) — Infinite-horizon world generation with real-time moving control.
+ [**Oasis**](https://oasis-model.github.io) (Decart, 2024) — Transformer-based real-time interactive world generation. [[GitHub]](https://github.com/etched-ai/open-oasis)
+ [**Genie**](https://arxiv.org/abs/2402.15391) (ICML, 2024) — Latent action discovery; generative interactive environment from video.

### Text-Interactive Worlds

+ [**From Word to World**](https://arxiv.org/abs/2512.18832) (ACL, 2026) — Evaluating LLMs as implicit text-based world models. [[GitHub]](https://github.com/X1AOX1A/Word2World)
+ [**WALL-E 2.0**](https://arxiv.org/abs/2504.15785) (NeurIPS, 2025) — Neuro-symbolic world alignment for text-world planning. [[GitHub]](https://github.com/elated-sawyer/WALL-E)
+ [**WALL-E 1.0**](https://arxiv.org/abs/2410.07484) (arXiv, 2024) — Rule-aligned world model for LLM agents in text environments. [[GitHub]](https://github.com/elated-sawyer/WALL-E)
+ [**RAP**](https://arxiv.org/abs/2305.14992) (EMNLP, 2023) — Reasoning with language model is planning with world model. [[GitHub]](https://github.com/Ber666/RAP)

### World Model Theory & Analysis

+ [**SafetyDrift**](https://arxiv.org/abs/2603.27148) (arXiv, 2026) — Predicting when AI agents cross the line before they actually do.
+ [**Neuro-Symbolic Synergy**](https://arxiv.org/abs/2602.10480) (arXiv, 2026) — Neuro-symbolic synergy for interactive world modeling. [[GitHub]](https://github.com/tianyi-lab/NeSyS)
+ [**Current Agents Fail to Leverage WM**](https://arxiv.org/abs/2601.03905) (ACL, 2026) — Negative evidence on agents using world models as foresight tools.
+ [**General Agents Contain World Models**](https://arxiv.org/abs/2506.01622) (ICML, 2025) — Theoretical result on world models in general agents.
+ [**RAFA**](https://arxiv.org/abs/2309.17382) (ICML, 2024) — LLM as world model for principled look-ahead planning in autonomous agents.
+ [**LAW**](https://arxiv.org/abs/2312.05230) (arXiv, 2023) — Language models, agent models, and world models: a framework for machine reasoning and planning.

## 🌐 Web & GUI

World models for web navigation, desktop/mobile GUI interaction, and computer-use agents.

### Web Navigation

+ [**WAC**](https://arxiv.org/abs/2602.15384) (arXiv, 2026) — World-model-augmented web agents with action correction.
+ [**WebWorld**](https://arxiv.org/abs/2602.14721) (arXiv, 2026) — Large-scale LLM-based world model for web agent training.
+ [**DynaWeb**](https://arxiv.org/abs/2601.22149) (arXiv, 2026) — Model-based reinforcement learning of web agents.
+ [**WebWM**](https://arxiv.org/abs/2512.23676) (arXiv, 2025) — Web World Models; semantic HTML delta prediction.
+ [**WebSynthesis**](https://arxiv.org/abs/2507.04370) (arXiv, 2025) — World-model-guided MCTS for efficient WebUI-trajectory synthesis. [[GitHub]](https://github.com/LucusFigoGao/WebSynthesis)
+ [**WebEvolver**](https://arxiv.org/abs/2504.21024) (EMNLP, 2025) — Enhancing web agent self-improvement with coevolving world model. [[GitHub]](https://github.com/Tencent/SelfEvolvingAgent)
+ [**WebDreamer**](https://arxiv.org/abs/2411.06559) (TMLR, 2025) — LLM web state simulation + tree search for planning. [[GitHub]](https://github.com/OSU-NLP-Group/WebDreamer)
+ [**WMA Web Agent**](https://arxiv.org/abs/2410.13232) (ICLR, 2025) — Learning and leveraging environment dynamics in web navigation. [[GitHub]](https://github.com/kyle8581/WMA-Agents)

### GUI — Desktop & Computer-Use

+ [**Computer-Using World Model**](https://arxiv.org/abs/2602.17365) (arXiv, 2026) — LLM-based UI state delta prediction for action search.
+ [**SafePred**](https://arxiv.org/abs/2602.01725) (arXiv, 2026) — Predictive guardrail for computer-using agents via world models. [[GitHub]](https://github.com/YurunChen/SafePred)
+ [**R-WoM**](https://arxiv.org/abs/2510.11892) (ICLR, 2026) — Retrieval-augmented world model for computer-use agents.
+ [**NeuralOS**](https://arxiv.org/abs/2507.08800) (ICLR, 2026) — RNN + pixel diffusion for operating system simulation. [[GitHub]](https://github.com/yuntian-group/neural-os)

### GUI — Mobile

+ [**Code2World**](https://arxiv.org/abs/2602.09856) (arXiv, 2026) — GUI world model via renderable code generation. [[GitHub]](https://github.com/AMAP-ML/Code2World)
+ [**GEBench**](https://arxiv.org/abs/2602.09007) (arXiv, 2026) — Benchmarking image generation models as GUI environments. [[GitHub]](https://github.com/stepfun-ai/GEBench)
+ [**gWorld**](https://arxiv.org/abs/2602.01576) (arXiv, 2026) — Generative visual code for mobile world modeling. [[GitHub]](https://github.com/trillion-labs/gWorld)
+ [**MobileDreamer**](https://arxiv.org/abs/2601.04035) (ICLR, 2026) — Generative sketch world model for GUI agents.
+ [**ViMo**](https://arxiv.org/abs/2504.13936) (ICLR, 2026) — Generative visual GUI world model for app agents.
+ [**MobileWorldBench**](https://arxiv.org/abs/2512.14014) (COLM, 2025) — Towards semantic world modeling for mobile agents. [[GitHub]](https://github.com/jacklishufan/MobileWorld)
+ [**UISim**](https://arxiv.org/abs/2509.21733) (NeurIPS Workshop, 2025) — Interactive image-based UI simulator for dynamic mobile environments.
+ [**ASL**](https://arxiv.org/abs/2506.17697) (arXiv, 2025) — Action semantics learning via UI state-transition estimation for mobile app agents.

### Cross-Domain GUI/Web

+ [**World Craft**](https://arxiv.org/abs/2601.09150) (arXiv, 2026) — Agentic framework to create visualizable worlds via text. [[GitHub]](https://github.com/HerzogFL/World-Craft)
+ [**EvoAgent**](https://arxiv.org/abs/2502.05907) (ICLR, 2026) — Self-evolving agent with continual world model for long-horizon tasks. [[GitHub]](https://github.com/fengtt42/EvoAgent)
+ [**VAGEN**](https://arxiv.org/abs/2510.16907) (NeurIPS, 2025) — Reinforcing world model reasoning for multi-turn VLM agents. [[GitHub]](https://github.com/mll-lab-nu/VAGEN)
+ [**SimuRA**](https://arxiv.org/abs/2507.23773) (arXiv, 2025) — World-model-driven simulative reasoning architecture for general goal-oriented agents.
+ [**WorldGPT**](https://doi.org/10.1145/3664647.3681488) (ACM MM, 2024) — Empowering LLMs as multimodal world models. [[GitHub]](https://github.com/DCDmllm/WorldGPT)

## 🔧 Tool Use

World models for tool-augmented agents, API interaction, and workflow systems.

+ [**Agent World Model**](https://arxiv.org/abs/2602.10090) (arXiv, 2026) — Infinity synthetic environments for agentic RL. [[GitHub]](https://github.com/Snowflake-Labs/agent-world-model)
+ [**RWML**](https://arxiv.org/abs/2602.05842) (arXiv, 2026) — Reinforcement world model learning for LLM-based agents.
+ [**World of Workflows**](https://arxiv.org/abs/2601.22130) (arXiv, 2026) — Benchmark for bringing world models to enterprise systems. [[GitHub]](https://github.com/Skyfall-Research/world-of-workflows)
+ [**GTM**](https://arxiv.org/abs/2512.04535) (arXiv, 2025) — Simulating the world of tools for AI agents.
+ [**LLMs as Simulators**](https://arxiv.org/abs/2510.14969) (arXiv, 2025) — LLMs as scalable, general-purpose simulators for evolving digital agent training. [[GitHub]](https://github.com/WadeYin9712/UI-Simulator)
+ [**ToolRM**](https://arxiv.org/abs/2509.11963) (arXiv, 2025) — Outcome reward models for tool-calling LLMs.
+ [**DyMo**](https://arxiv.org/abs/2506.02918) (arXiv, 2025) — Augments LLMs with dynamics modelling for state prediction and self-verification in tool-calling agents.

## 💻 Code

World models for code generation, software engineering, and executable environment modeling.

+ [**SWE-World**](https://arxiv.org/abs/2602.03419) (arXiv, 2026) — Building software engineering agents in docker-free environments. [[GitHub]](https://github.com/RUCAIBox/SWE-World)
+ [**CWM**](https://arxiv.org/abs/2510.02387) (arXiv, 2025) — Open-weights LLM for research on code generation with world models. [[GitHub]](https://github.com/facebookresearch/cwm)
+ [**Code World Models (GIF-MCTS)**](https://arxiv.org/abs/2405.15383) (NeurIPS, 2024) — Generating code world models with LLMs guided by Monte Carlo tree search. [[GitHub]](https://github.com/nicoladainese96/code-world-models)
+ [**WorldCoder**](https://arxiv.org/abs/2402.12275) (NeurIPS, 2024) — Model-based LLM agent that builds world models by writing code and interacting with the environment. [[GitHub]](https://github.com/haotang1995/WorldCoder)

## 🏋️ Benchmarks & Environments

Representative interactive benchmarks and environments for evaluating world models in digital-agent settings, organized by domain. Descriptions highlight the primary predictive difficulty each environment introduces.

### Games

+ [**GameWorld**](https://arxiv.org/abs/2604.07429) (arXiv, 2026) — Browser-game agent benchmark with state-verifiable outcome metrics across 34 games and 170 tasks. [[GitHub]](https://github.com/gameworld-project/gameworld)
+ [**PlanCraft**](https://arxiv.org/abs/2412.21033) (COLM, 2025) — Structured planning over branching crafting dependencies. [[GitHub]](https://github.com/gautierdag/plancraft)
+ [**Minigrid**](https://arxiv.org/abs/2306.13831) (NeurIPS, 2023) — Compact planning under partially observable gridworld dynamics. [[GitHub]](https://github.com/Farama-Foundation/Minigrid)
+ [**SMACv2**](https://arxiv.org/abs/2212.07489) (NeurIPS, 2023) — Harder stochastic branching and robust multi-agent rollout. [[GitHub]](https://github.com/oxwhirl/smacv2)
+ [**MineDojo**](https://arxiv.org/abs/2206.08853) (NeurIPS, 2022) — Open-ended embodied agents with internet-scale knowledge in Minecraft. [[GitHub]](https://github.com/MineDojo/MineDojo)
+ [**ScienceWorld**](https://arxiv.org/abs/2203.07540) (EMNLP, 2022) — Causal reasoning over structured textual world state. [[GitHub]](https://github.com/allenai/ScienceWorld)
+ [**Crafter**](https://arxiv.org/abs/2109.06780) (ICLR, 2022) — Persistent survival-state dynamics and open-ended progression. [[GitHub]](https://github.com/danijar/crafter)
+ [**MiniHack**](https://arxiv.org/abs/2109.13202) (NeurIPS, 2021) — Modular long-horizon transition modeling across diverse tasks. [[GitHub]](https://github.com/NetHack-LE/minihack)
+ [**ALFWorld**](https://arxiv.org/abs/2010.03768) (ICLR, 2021) — Multi-step symbolic planning with delayed consequences. [[GitHub]](https://github.com/alfworld/alfworld)
+ [**NetHack**](https://arxiv.org/abs/2006.13760) (NeurIPS, 2020) — Partial observability and long-horizon symbolic-visual state tracking. [[GitHub]](https://github.com/NetHack-LE/nle)
+ [**Procgen**](https://arxiv.org/abs/1912.01588) (ICML, 2020) — Generalization under procedurally varying dynamics. [[GitHub]](https://github.com/openai/procgen)
+ [**Atari 100k**](https://arxiv.org/abs/1903.00374) (arXiv, 2019) — Model-based RL benchmark with 100k interaction steps per game. [[GitHub]](https://github.com/YeWR/EfficientZero)
+ [**SMAC**](https://arxiv.org/abs/1902.04043) (AAMAS, 2019) — Multi-agent state evolution and coordination-sensitive rollout. [[GitHub]](https://github.com/oxwhirl/smac)
+ [**BabyAI**](https://arxiv.org/abs/1810.08272) (ICLR, 2019) — Grounded language-conditioned planning and hidden-state tracking. [[GitHub]](https://github.com/mila-iqia/babyai)
+ [**TextWorld**](https://arxiv.org/abs/1806.11532) (CGW@IJCAI, 2018) — Long-horizon symbolic transitions and command-conditioned dynamics. [[GitHub]](https://github.com/Microsoft/TextWorld)
+ [**Atari 200M**](https://doi.org/10.1038/nature14236) (Nature, 2015) — Visual dynamics modeling under long-horizon control. [[GitHub]](https://github.com/Farama-Foundation/Arcade-Learning-Environment)

### Web & GUI

+ [**GUI-360°**](https://arxiv.org/abs/2511.04307) (ICLR, 2026) — Comprehensive dataset and benchmark for computer-using agents. [[GitHub]](https://github.com/2020-qqtcg/GUI-360)
+ [**MobileWorld**](https://arxiv.org/abs/2512.19432) (arXiv, 2025) — Mobile workflows with user interaction and MCP-augmented tasks. [[GitHub]](https://github.com/Tongyi-MAI/MobileWorld)
+ [**TheAgentCompany**](https://arxiv.org/abs/2412.14161) (NeurIPS, 2025) — Enterprise-style browser workflows and hidden application logic. [[GitHub]](https://github.com/TheAgentCompany/TheAgentCompany)
+ [**WindowsAgentArena**](https://arxiv.org/abs/2409.08264) (ICML, 2025) — Real multi-app state coordination under desktop interaction. [[GitHub]](https://github.com/microsoft/WindowsAgentArena)
+ [**AndroidWorld**](https://arxiv.org/abs/2405.14573) (ICLR, 2025) — Dynamic mobile UI transitions under touch interaction. [[GitHub]](https://github.com/google-research/android_world)
+ [**OSWorld**](https://arxiv.org/abs/2404.07972) (NeurIPS, 2024) — Open-ended computer-use transitions across apps and OS state. [[GitHub]](https://github.com/xlang-ai/OSWorld)
+ [**OmniACT**](https://arxiv.org/abs/2402.17553) (ECCV, 2024) — Executable computer-use actions with cross-platform UI effects. [[Website]](https://huggingface.co/datasets/Writer/omniact)
+ [**VisualWebArena**](https://arxiv.org/abs/2401.13649) (ACL, 2024) — Multimodal web-state prediction with visual-semantic grounding. [[GitHub]](https://github.com/web-arena-x/visualwebarena)
+ [**WebArena**](https://arxiv.org/abs/2307.13854) (ICLR, 2024) — Semantic UI transitions under realistic web workflows. [[GitHub]](https://github.com/web-arena-x/webarena)
+ [**WebShop**](https://arxiv.org/abs/2207.01206) (NeurIPS, 2022) — Workflow branching and latent task progress in shopping interfaces. [[GitHub]](https://github.com/princeton-nlp/WebShop)
+ [**MiniWoB++**](https://arxiv.org/abs/1802.08802) (ICLR, 2018) — Short-horizon UI changes with typed web actions. [[GitHub]](https://github.com/Farama-Foundation/miniwob-plusplus)
+ [**World of Bits**](https://proceedings.mlr.press/v70/shi17a.html) (ICML, 2017) — Open-domain web-state transitions under interface interaction.

### Tool Use

+ [**τ-Voice**](https://arxiv.org/abs/2603.13686) (arXiv, 2026) — Voice-grounded tool use with multimodal state tracking.
+ [**AgentWorldModel-1K**](https://arxiv.org/abs/2602.10090) (arXiv, 2026) — Synthetic external-state modeling in executable SQL-MCP ecosystems. [[GitHub]](https://github.com/Snowflake-Labs/agent-world-model)
+ [**World of Workflows**](https://arxiv.org/abs/2601.22130) (arXiv, 2026) — Enterprise workflow transitions with hidden service-side dependencies. [[GitHub]](https://github.com/Skyfall-Research/world-of-workflows)
+ [**MCP-Universe**](https://arxiv.org/abs/2508.14704) (arXiv, 2025) — Multi-server tool coordination and external-state mutation. [[GitHub]](https://github.com/SalesforceAIResearch/MCP-Universe)
+ [**τ²-bench**](https://arxiv.org/abs/2506.07982) (arXiv, 2025) — Multi-party control and persistent tool-state consistency. [[GitHub]](https://github.com/sierra-research/tau2-bench)
+ [**BFCL v3**](https://gorilla.cs.berkeley.edu/blogs/13_bfcl_v3_multi_turn.html) (ICML, 2025) — Multi-turn function-call validity and argument correctness. [[GitHub]](https://github.com/ShishirPatil/gorilla)
+ [**τ-bench**](https://arxiv.org/abs/2406.12045) (ICLR, 2025) — Typed tool-state updates in tool-agent-user interaction. [[GitHub]](https://github.com/sierra-research/tau-bench)

### Code

+ [**SWE-World**](https://arxiv.org/abs/2602.03419) (arXiv, 2026) — Software-engineering transitions with surrogate environment feedback. [[GitHub]](https://github.com/RUCAIBox/SWE-World)
+ [**SWE-bench**](https://arxiv.org/abs/2310.06770) (ICLR, 2024) — Repository-scale state evolution under code edits and tests. [[GitHub]](https://github.com/swe-bench/SWE-bench)
+ [**InterCode**](https://arxiv.org/abs/2306.14898) (NeurIPS, 2023) — Interactive coding dynamics with stepwise execution feedback. [[GitHub]](https://github.com/princeton-nlp/intercode)
+ [**HumanEval**](https://arxiv.org/abs/2107.03374) (arXiv, 2021) — Executable outcome prediction for code generation. [[GitHub]](https://github.com/openai/human-eval)

### Multi-Domain

+ [**AgentGym**](https://arxiv.org/abs/2406.04151) (ACL, 2025) — Transfer of predictive abstractions across heterogeneous environments. [[GitHub]](https://github.com/WooooDyy/AgentGym)
+ [**AgentBench**](https://arxiv.org/abs/2308.03688) (ICLR, 2024) — Cross-domain comparison of agent behavior and predictive utility. [[GitHub]](https://github.com/THUDM/AgentBench)

## 📊 Datasets

Datasets used to train and evaluate digital-agent world models, organized by supervision type and the role each category plays in the learning pipeline.

| Supervision Type | Representative Datasets | Role in World Models |
|:-----------------|:-----------------------|:---------------------|
| **Action-conditioned transitions** | [Mind2Web](https://arxiv.org/abs/2306.06070), [Android in the Wild](https://arxiv.org/abs/2307.10088), [Android Control](https://arxiv.org/abs/2406.03679), [MobileWorldBench](https://arxiv.org/abs/2512.14014) | Semantic UI dynamics, DOM/screenshot deltas, action-conditioned next-state prediction |
| **Trajectory demonstrations and behavioral priors** | [GUI-360°](https://arxiv.org/abs/2511.04307), [WebLINX](https://arxiv.org/abs/2402.05930), [AMEX](https://arxiv.org/abs/2407.17490), [Atari-HEAD](https://arxiv.org/abs/1903.06754), [MineRL](https://arxiv.org/abs/1907.13440), [VPT](https://arxiv.org/abs/2206.11795), [TextWorld](https://arxiv.org/abs/1806.11532), [ALFWorld](https://arxiv.org/abs/2010.03768), [ScienceWorld](https://arxiv.org/abs/2203.07540) | Planning priors, long-range workflow structure, reducing long-horizon compounding error |
| **Execution-grounded supervision** | [APPS](https://arxiv.org/abs/2105.09938), [MBPP](https://arxiv.org/abs/2108.07732), [CodeXGLUE](https://arxiv.org/abs/2102.04664), [HumanEval](https://arxiv.org/abs/2107.03374) | Constraint-aware prediction, execution-grounded transition modeling, code-world semantics |
| **Tool interaction traces** | [ToolBench](https://arxiv.org/abs/2307.16789), [ToolAlpaca](https://arxiv.org/abs/2306.05301), [API-Bank](https://arxiv.org/abs/2304.08244) | Schema-constrained actions, external system dynamics, multi-step tool workflows |
| **Large-scale priors and foundation corpora** | [The Stack](https://arxiv.org/abs/2211.15533), [StarCoder Data](https://arxiv.org/abs/2305.06161) | Reusable semantic priors, representation quality, cross-domain generalization |
| **Synthetic and simulated trajectories** | [WebWorld](https://arxiv.org/abs/2602.14721), [Agent World Model](https://arxiv.org/abs/2602.10090) | Expanding branching coverage, surfacing rare or risky states, planning robustness |

## 📚 Related Surveys

Surveys on world models, autonomous agents, and related topics that provide complementary perspectives to our work.

+ [**Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond**](https://arxiv.org/abs/2604.22748) (arXiv, 2026)
+ [**Learning to Model the World: A Survey of World Models in Artificial Intelligence**](https://www.preprints.org/manuscript/202603.0739) (Preprints, 2026)
+ [**Towards Generalist Embodied AI: A Survey on World Models for VLA Agents**](https://www.techrxiv.org/doi/10.36227/techrxiv.176948355.54623875) (TechRxiv, 2026)
+ [**A Step Toward World Models: A Survey on Robotic Manipulation**](https://arxiv.org/abs/2511.02097) (arXiv, 2025)
+ [**From Masks to Worlds: A Hitchhiker's Guide to World Models**](https://arxiv.org/abs/2510.20668) (arXiv, 2025)
+ [**A Comprehensive Survey on World Models for Embodied AI**](https://arxiv.org/abs/2510.16732) (arXiv, 2025)
+ [**3D and 4D World Modeling: A Survey**](https://arxiv.org/abs/2509.07996) (arXiv, 2025)
+ [**A Survey: Learning Embodied Intelligence from Physical Simulators and World Models**](https://arxiv.org/abs/2507.00917) (arXiv, 2025)
+ [**The Role of World Models in Shaping Autonomous Driving: A Comprehensive Survey**](https://arxiv.org/abs/2502.10498) (arXiv, 2025)
+ [**A Survey of World Models for Autonomous Driving**](https://arxiv.org/abs/2501.11260) (arXiv, 2025)
+ [**Understanding World or Predicting Future? A Comprehensive Survey of World Models**](https://arxiv.org/abs/2411.14499) (ACM CSUR, 2025)
+ [**Is Sora a World Simulator? A Comprehensive Survey on General World Models and Beyond**](https://arxiv.org/abs/2405.03520) (arXiv, 2024)
+ [**A Survey on Large Language Model Based Autonomous Agents**](https://arxiv.org/abs/2308.11432) (Frontiers, 2024)

## 🤝 Contributing

**We welcome contributions!** This project is actively maintained. If you know a paper or benchmark that should be listed, please [open an issue](https://github.com/Darwin-Agent/awesome-world-models-for-digital-agents/issues/new?template=add_paper.yml) or submit a pull request.

When submitting, please include:
- **Title** and **paper URL** (arXiv or publication link)
- **Venue** and **year**
- **Target section** (Games / Web & GUI / Tool Use / Code / Benchmarks)
- **One-line summary** of the contribution
- **Code URL** (optional, for GitHub link)
- **Suggested README entry** (optional)

## 📖 Citation

If you find this resource useful, please cite our survey:

```bibtex
@article{chen2026digitalagentsworldmodels,
  title         = {Digital Agents Meet World Models: A Survey},
  author        = {Yuhan Chen and Pengzhi Gao and Qinzhuo Wu and Kun Huang and Yike Liu and Heng Qu and Kun Shao and Jian Luan},
  year          = {2026},
  url           = {https://github.com/Darwin-Agent/awesome-world-models-for-digital-agents}
}
```

awesome-world-models-for-digital-agents

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README