An Open-Ended Realistic Simulator for Agents in Physical and Social Worlds (simworld.org)

A new AI simulation platform, SimWorld, has been announced, offering an open-ended realistic simulation environment conducive to both physical and social dynamics. The platform features a modular procedural generation system that creates complex environments with a focus on realism, alongside advanced language-based world editing capabilities. This allows users and agents to modify scenes using natural-language commands, enhancing interactive experiences. Significantly, SimWorld presents a rich interface for large language models (LLMs) and vision-language models (VLMs), enabling multimodal observations and grounded actions across various abstraction levels. This supports diverse reasoning scenarios that facilitate systematic training and evaluation of agents in complex environments. With capabilities for open-vocabulary action space and an effective action planning system, SimWorld positions itself as a versatile tool for the AI/ML community, aiming to advance research in agent behavior and reasoning in both physical and social contexts.