SimWorld

An Open-ended Realistic Simulator for Autonomous Agents in Physical and Social Worlds

🌍 Open-ended Realistic Simulation

🤖 Rich LLM/VLM Agent Interface

💡 Diverse Reasoning Scenarios

Simulator Comparison

Simulator Open-ended Realistic Simulation Rich LLM/VLM Agent Interface Diverse Reasoning Scenarios
Simulation Realism Procedural Generation Language Control Open Vocabulary Action Space High-level Control Low-level Control Social Reasoning Physical Reasoning
SimWorld ★★★
AI2-THOR ★★ - - - -
Genesis ★★★ - - - -
VirtualCommunity ★★ - - -
Mindcraft - - -
Minedojo - - - - -
MetaUrban ★★ - - - -
EmbodiedCity ★★★ - - - - - -
CARLA ★★★ - - - - -
GRUtopia ★★ - - - - -
OmniGibson ★★ - - - -
Habitat 3.0 ★★ - - - - -
UnrealZoo ★★★ - - - - -

Open-ended Realistic Simulation

Procedural Scene Generation

SimWorld’s procedural generation system uses a modular, extensible pipeline with three stages: road generation, building generation, and street-element generation, each adding more structural and visual detail.

Various Environments

SimWorld offers a broad spectrum of meticulously designed environments, enabling diverse world-building and scenario development.

Loading video...

Physical and Social Dynamics

SimWorld simulates realistic physical, environmental, and social dynamics that shape the behavior of agents and the world around them.

Loading video...

Physical laws (e.g., gravity, momentum)

Loading video...

Lighting, weather, time of day

Loading video...

Traffic System

Language-based World Editing

Beyond static and procedurally generated maps, SimWorld supports open-ended, language-based world editing, allowing users and agents to create, modify, and compose scenes on the fly with natural-language commands.

Loading video...

“Generate several buildings that can fill the current empty block.”

Loading video...

“Generate a motorcycle and put it in the middle of the road.”

Loading video...

“Replace the buildings to make the overall style more consistent.”

Rich LLM/VLM Agent Interface

SimWorld provides a comprehensive interface for LLM/VLM agents with rich observation modalities and diverse action capabilities, enabling agents to perceive and interact with the environment in a natural and intuitive manner.

Observation Space

The simulator provides diverse observations including visual sensors (RGB, depth, segmentation), scene graph and GPS information (global and local maps).

RGB Observation

RGB

Depth Observation

Depth

Segmentation Observation

Segmentation

Scene Graph

Scene Graph

Global Map

Global Map

Local Map

Local Map

Open-Vocabulary Action Space

SimWorld supports an open-vocabulary action space that accepts natural language commands, which are then decomposed by a built-in action planner into sequences of low-level primitive actions.

Vehicle Action

Driving vehicles in realistic traffic

Argue Action

Natural social interaction between agents

Human-Robot Interaction

Human–robot collaboration in shared spaces

Pickup Action

Picking up and delivering objects

Pickup Action 2

Fine-grained object manipulation

Point Action

Pointing and gesturing to ground language

Diverse Reasoning Scenarios

Enable agents to perform complex reasoning and coordinated behaviors across diverse physical and social contexts.

Loading video...

Low-level motion control while avoiding obstacles.

Loading video...

Multimodal instruction-following navigation with visual hints.

Loading video...

Deliver food across the city, completing orders to earn money.

Research in SimWorld

SimWorld Technical Report
White Paper

SimWorld: An Open-ended Realistic Simulator for Autonomous Agents in Physical and Social Worlds

Authors: Jiawei Ren*, Yan Zhuang*, Xiaokang Ye*, Lingjun Mao, Xuhong He, Jianzhi Shen, Mrinaal Dogra, Yiming Liang, Ruixuan Zhang, Tianai Yue, Yiqing Yang, Eric Liu, Ryan Wu, Kevin Benavente, Rajiv Mandya Nagaraju, Muhammad Faayez, Xiyan Zhang, Dhruv Vivek Sharma, Xianrui Zhong, Ziqiao Ma, Tianmin Shu†, Zhiting Hu†, Lianhui Qin†

Paper
DeliveryBench

DeliveryBench: Can Agents Earn Profit in Real World?

Authors: Lingjun Mao, Jiawei Ren, Kun Zhou, Jixuan Chen, Ziqiao Ma, Lianhui Qin

Coming soon!
SimWorld NeurIPS Spotlight

SimWorld: An Open-ended Simulator for Agents in Physical and Social Worlds

Authors: Xiaokang Ye*, Jiawei Ren*, Yan Zhuang, Xuhong He, Yiming Liang, Yiqing Yang, Xianrui Zhong, Mrinaal Dogra, Eric Liu, Kevin Benavente, Rajiv Mandya Nagaraju, Dhruv Vivek Sharma, Ziqiao Ma, Tianmin Shu†, Zhiting Hu†, Lianhui Qin†

Venue: NeurIPS 2025 (Spotlight 🏆)

Paper
SimWorld NeurIPS Paper

SimWorld-Robotics: Synthesizing Photorealistic and Dynamic Urban Environments for Multimodal Robot Navigation and Collaboration

Authors: Yan Zhuang, Jiawei Ren*, Xiaokang Ye*, Jianzhi Shen, Ruixuan Zhang, Tianai Yue, Muhammad Faayez, Xuhong He, Ziqiao Ma, Lianhui Qin†, Zhiting Hu†, Tianmin Shu†

Venue: NeurIPS 2025

Paper Repo Website
SimWorld CVPR Demo

SimWorld: A World Simulator for Scaling Photorealistic Multi-Agent Interactions

Authors: Yan Zhuang*, Jiawei Ren*, Xiaokang Ye*, Xuhong He, Zijun Gao, Ryan Wu, Mrinaal Dogra, Cassie Zhang, Kai Kim, Bertt Wolfinger, Ziqiao Ma, Tianmin Shu†, Zhiting Hu†, Lianhui Qin†

Venue: CVPR 2025 Demo

Organizations

UCSD
JHU