A reference catalog for embodied AI
The reference catalog of world models and simulation environments.
Browse, search, and compare the world model ecosystem — the models, the simulation environments they run in, and the infrastructure used to train them.
Genie 2
Google DeepMind
Foundation world model that generates playable, action-controllable 3D environments from a single image.
1 relationships
NVIDIA Cosmos
NVIDIA
World foundation model platform for physical AI, with open diffusion and autoregressive video world models.
4 relationships
V-JEPA 2
Meta AI (FAIR)
Self-supervised video joint-embedding predictive world model for understanding and planning.
1 relationships
DreamerV3
Google DeepMind / Danijar Hafner
General model-based RL agent that learns a world model (RSSM) and plans in imagination across domains.
2 relationships
World Labs
World Labs
Large World Models that generate persistent, explorable 3D worlds from a single image or prompt.
GAIA-2
Wayve
Controllable video generative world model for autonomous driving scenario synthesis.
2 relationships
Oasis
Decart / Etched
Real-time, fully generated playable world model — an interactive game rendered frame-by-frame by a transformer.
MuJoCo
Google DeepMind
Fast, accurate rigid-body physics engine for robotics and RL; MJX adds GPU-batched, differentiable simulation.
3 relationships
NVIDIA Isaac Sim
NVIDIA
Omniverse/USD robotics simulator with photorealistic RTX rendering and physically accurate sensor simulation.
4 relationships
Genesis
Genesis Embodied AI
Generative, GPU-accelerated physics platform unifying rigid, soft, fluid, and differentiable simulation.
2 relationships
AI Habitat
Meta AI (FAIR)
High-performance simulator for embodied AI in photorealistic 3D scans, with human-robot interaction (Habitat 3.0).
ManiSkill
UC San Diego (Hao Su Lab)
GPU-parallelized robotic manipulation simulation and benchmark built on SAPIEN.
1 relationships
Antioch
Antioch
Full-stack cloud simulation platform to test and deploy autonomous systems, with parallel cloud runs and an agent.
1 relationships
JAX
Composable transforms of NumPy programs — autodiff, JIT to XLA, and vmap/pmap — the backbone of large-scale world-model training on TPU/GPU.
1 relationships
PyTorch
Meta / Linux Foundation
The dominant deep-learning framework; eager autograd plus torch.compile, FSDP, and a broad ecosystem for training video and world models.
3 relationships
Ray / RLlib
Anyscale
Distributed compute framework for Python; RLlib provides scalable reinforcement-learning training used to learn policies inside world models and simulators.
NVIDIA Warp
NVIDIA
Python framework for writing differentiable GPU kernels; powers high-throughput, gradient-capable physics used in modern simulation and world-model training.
1 relationships
NVIDIA Isaac Lab
NVIDIA
Open GPU-accelerated robot-learning framework built on Isaac Sim; standardizes massively-parallel RL and imitation training environments.
2 relationships