World models

Research

World-model resources for physical AI and synthetic-data workflows.

Research

Repository for our papers: Robotic World Model: A Neural Network Simulator for Robust Policy Optimization in Robotics and Uncertainty-Aware Robotic World Model Makes Offline Model-Based…

Research

DayDreamer provides world-model methods for physical robot learning.

Research

A video-action model research project for robotic control.

Research

DreamDojo studies robot world models learned from human video data.

Research

Official Implementation for the paper "SR-AIF: Solving Sparse-Reward Robotic Tasks from Pixels with Active Inference and World Models"

Research

A robot policy project using world-model representations.

Research

world modeling challenge for humanoid robots

Research

A unified framework for training, fine-tuning, and evaluating World Action Models

Research

ICCV 2025 | TesserAct: Learning 4D Embodied World Models

Research

Code for RISE: Self-Improving Robot Policy with Compositional World Model

Research

A world-model-enhanced VLA for steerable dexterous manipulation, along with optimized training infra.

Research

Lightweight implementation of Robotic World Model (RWM) and Uncertainty-Aware Robotic World Model (RWM-U)

Research

ACOT-VLA-WM is an advanced Vision-Language-Action robotics framework that uses a predictive world model to generate visual subgoals, improving long-horizon robotic manipulation accuracy and…

Research

Aether: Geometric-Aware Unified World Modeling

Research

C³-JEPA studies a multi-view world model for underwater ROV salvage.

Research

Official code and data from DexWM ("World Models Can Leverage Human Videos for Dexterous Manipulation").

Research

DreamX-Phi provides an action-conditioned video world model for robotic manipulation.

Research

Ego-World JEPA studies ego/world latent factorization for planning on PushT.

Research

Official Implementation of "EVA: Aligning Video World Models with Executable Robot Actions via Inverse Dynamics Rewards"

Research

EVEWorld studies physical-evolution supervision and temporal consistency in embodied world models.

Research

Towards Scalable Gaussian World Models for Robotic Manipulation

Research

GenRL: Multimodal-foundation world models enable grounding language and video prompts into embodied domains, by turning them into sequences of latent world model states. Latent state sequences can…

Research

A Roadmap to Build World Models for Robot Policy Evaluation