Minimal Local Simulation Foundations for LLM- and VLM-Driven Agents in 2D and 3D Environments
Researchers have created two minimal simulation foundations for large language models (LLMs) and vision-language models (VLMs). The 2D and 3D environments, called SD-AgentFoundry-2D and SD-AgentFoundry-3D, allow locally hosted agents to interact with their surroundings. These platforms are designed for education and rapid prototyping, can run on ordinary computers, and are intentionally left open for modification.
Researchers have created two minimal simulation foundations for large language models (LLMs) and vision-language models (VLMs). The 2D and 3D environments, called SD-AgentFoundry-2D and SD-AgentFoundry-3D, allow locally hosted agents to interact with their surroundings. These platforms are designed for education and rapid prototyping, can run on ordinary computers, and are intentionally left open for modification.
---
Why it matters: This work matters because it provides accessible starting points for researchers and developers to learn about generative social simulation and build domain-specific extensions. The ability to run these simulations locally on various operating systems makes them more approachable and efficient to study and modify.
Source: https://arxiv.org/abs/2608.22833
This article was originally published at: https://arxiv.org/abs/2608.22833