世界模型论文趋势 2023–2026

主要结论

• 世界模型论文从 2023 年的 23 篇增长到 2026 年的 250 篇，三年间扩大到约 11 倍。
• NeurIPS 是发世界模型论文最多的会议，2023–2026 年共 159 篇，占四个会议总量的 37%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 34 篇。
• 在「前沿系统」的 4 个主题里，它的 2026 年论文数排第 4；同组增长最快的AI 智能体是 12 倍，世界模型是 11 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026A Hierarchical World Model for Driving
• ICML 2026A Deep Learning Model of Mental Rotation Informed by Interactive VR Experiments
• ICLR 2026Astra: General Interactive World Model with Autoregressive Denoising
• CVPR 2026Active Intelligence in Video Avatars via Closed-loop World Modeling
• NeurIPS 2025COME: Adding Scene-Centric Forecasting Control to Occupancy World Model
• ICML 2025A Causal World Model Underlying Next Token Prediction: Exploring GPT in a Controlled Environment
• ICLR 2025AdaWM: Adaptive World Model based Planning for Autonomous Driving
• CVPR 2025DIO: Decomposable Implicit 4D Occupancy-Flow World Model
• NeurIPS 2024A Data-Centric Perspective on Evaluating Machine Learning Models for Tabular Data
• ICML 20243D-VLA: A 3D Vision-Language-Action Generative World Model
• ICLR 2024Copilot4D: Learning Unsupervised World Models for Autonomous Driving via Discrete Diffusion
• CVPR 2024DriveWorld: 4D Pre-trained Scene Understanding via World Models for Autonomous Driving
• NeurIPS 2023Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models
• ICML 2023Do Embodied Agents Dream of Pixelated Sheep: Embodied Decision Making using Language Guided World Modelling
• ICLR 2023Dynamic Update-to-Data Ratio: Minimizing World Model Overfitting
• NeurIPS 2026A World Model of Radiologist Reading for Medical Image Representation Learning
• ICML 2026Behavior-Invariant Task Representation Learning with Transformer-based World Models for Offline Meta-Reinforcement Learning
• ICLR 2026Building spatial world models from sparse transitional episodic memories
• CVPR 2026Chain of World World Model Thinking in Latent Motion
• NeurIPS 2025DMWM: Dual-Mind World Model with Long-Term Imagination

全部世界模型论文

共 159 篇，第 2 / 4 页

51. NeurIPS 2026LoopNav: Benchmarking Spatial Consistency in World Models
52. NeurIPS 2026MapShift: Controlled Post-Intervention Evaluation for Embodied World Models
53. NeurIPS 2026Masked Visual Actions for Unified World Modeling
54. NeurIPS 2026Medical LLMs as Medical World Models: Unified Policy-Dynamics Learning with Test-Time Search
55. NeurIPS 2026MIRAGE: Mobile Agents with Implicit Reasoning and Generative World Models
56. NeurIPS 2026Next Forcing: Causal World Modeling with Multi-Chunk Prediction
57. NeurIPS 2026Next-Latent Prediction Transformers Learn Compact World Models
58. NeurIPS 2026ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow
59. NeurIPS 2026PanoWorld: Geometry-Consistent Panoramic Video World Modeling
60. NeurIPS 2026Perception for Action in Latent World Models
61. NeurIPS 2026Plan4D: Generative Plannable 4D Worlds
62. NeurIPS 2026Predicting What Changes: Causal Delta World Models for Risk-Aware LLM Agent Planning
63. NeurIPS 2026Predictive but Not Plannable: RC-aux for Latent World Models
64. NeurIPS 2026Q-ARVD: Quantizing Autoregressive Video Diffusion Models
65. NeurIPS 2026Reconstruction or Semantics? What Makes a Latent Space Useful for Robotic World Models
66. NeurIPS 2026ReGen: Agentic Video World Modeling with Synergized Reasoning and Generation
67. NeurIPS 2026RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models
68. NeurIPS 2026SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts
69. NeurIPS 2026Solaris: Building a Multiplayer Video World Model in Minecraft
70. NeurIPS 2026Sparse Planning in Visual World Models via Cost Gradients
71. NeurIPS 2026StarWM: Self-Supervised Trained Attention Routing for Robust World Models
72. NeurIPS 2026Stealthy World Model Manipulation via Data Poisoning
73. NeurIPS 2026SurgVista: Long-Horizon Surgical World Modeling with Plausible Instrument-Tissue Dynamics
74. NeurIPS 2026TabPrep: Closing the Feature Engineering Gap in Tabular Benchmarks
75. NeurIPS 2026Think Before You Generate: Active Panoramic Exploration for Text-to-3D Scenes
76. NeurIPS 2026ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model
77. NeurIPS 2026Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models
78. NeurIPS 2026Towards Financial World Modeling
79. NeurIPS 2026Trajectory Evaluation via Rollout-Free World Model for End-to-end Autonomous Driving
80. NeurIPS 2026Trimming the Long-Tail of Visual World Modeling Evaluation
81. NeurIPS 2026When does LeJEPA learn a World Model?
82. NeurIPS 2026World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry
83. NeurIPS 2026World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning
84. NeurIPS 2026WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes
85. NeurIPS 2026WorldPrism: 3D Consistency for Video World Models via Bidirectional Cross-Space Verification
86. NeurIPS 2025Agents Robust to Distribution Shifts Learn Causal World Models Even Under Mediation
87. NeurIPS 2025Bootstrap Off-policy with World Model
88. NeurIPS 2025COME: Adding Scene-Centric Forecasting Control to Occupancy World Model
89. NeurIPS 2025DMWM: Dual-Mind World Model with Long-Term Imagination
90. NeurIPS 2025DyMoDreamer: World Modeling with Dynamic Modulation
91. NeurIPS 2025Dyn-O: Building Structured World Models with Object-Centric Representations
92. NeurIPS 2025Dynamics-Aligned Latent Imagination in Contextual World Models for Zero-Shot Generalization
93. NeurIPS 2025DynamicVerse: A Physically-Aware Multimodal Framework for 4D World Modeling
94. NeurIPS 2025EDELINE: Enhancing Memory in Diffusion-based World Models via Linear-Time Sequence Modeling
95. NeurIPS 2025From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction
96. NeurIPS 2025Image as a World: Generating Interactive World from Single Image via Panoramic Video Generation
97. NeurIPS 2025Learning 3D Persistent Embodied World Models
98. NeurIPS 2025Learning and Planning Multi-Agent Tasks via an MoE-based World Model
99. NeurIPS 2025Learning Interactive World Model for Object-Centric Reinforcement Learning
100. NeurIPS 2025Learning World Models for Interactive Video Generation

常见问题

2023–2026 年四大会议一共有多少篇世界模型论文？

共 425 篇，按年份分别是 2023 年 23 篇、2024 年 46 篇、2025 年 106 篇、2026 年 250 篇。

NeurIPS 2026 有多少篇世界模型论文？

NeurIPS 2026 有 85 篇世界模型论文，上一届是 42 篇。

ICLR 2026 有多少篇世界模型论文？

ICLR 2026 有 56 篇世界模型论文，上一届是 24 篇。

ICML 2026 有多少篇世界模型论文？

ICML 2026 有 69 篇世界模型论文，上一届是 27 篇。

CVPR 2026 有多少篇世界模型论文？

CVPR 2026 有 40 篇世界模型论文，上一届是 13 篇。

哪个会议的世界模型论文最多？

NeurIPS，2023–2026 年共 159 篇。