Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

world model for agents

Imagine an agent, like a robot or game character, that builds a mental simulator of its world. This "world model" lets it predict what will happen if it takes an action, so it can plan better without trying everything in reality. Think of it as a private video game running in the agent's head, where it can test moves before committing.

The papers show that naive world models often fail: they confuse visually identical situations, ignore tiny action-driven changes, or break with unseen teammates. Researchers fix this by injecting hidden state from past history, forcing the model to recover which action was taken, adding 3D depth, or building worlds from code with persistent memory. Some even predict task progress instead of pixels. The common thread is that reliable world models must look beyond surface observations to capture true dynamics, generalization, and long-term consistency.

Papers
113
Released code
2
First seen
Aug 2026
Latest
Nov 2026

109 papers in the last two months, against 4 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1World model simulator works across 22 robot platforms, enables policy learning

    PrimeSim turns a pretrained video generator into an action-conditioned, multiview world model that simulates robot manipulation across embodiments at 10.1 FPS.

    Nov 2026
  2. 2Driving simulators follow planned paths more faithfully with overhead views

    BirdDreams jointly generates perspective and bird's-eye-view videos to create reactive driving simulations from a single image.

    Nov 2026
  3. 3AI models can forecast sales well yet misjudge price changes

    Forecast accuracy does not indicate how accurately a model predicts the effect of price changes.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

diffusion modelA type of generative model that learns to reverse a gradual noising process to create data.23 togethervision-language-action modelA robot policy model that maps visual observations and language instructions to actions, trained on demonstrations and often used for manipulation tasks.22 togetherflow matchingA generative training method that learns an ordinary differential equation mapping noise to data by matching velocity fields along probability paths.20 together

Who is working on it

National University of Singapore 8NVIDIA 7Peking University 5Nanjing University 4Amazon 4Tencent 4

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.