GRUtopia is a general-purpose research platform for embodied AGI, targeting researchers and developers in robotics and AI. It aims to alleviate data scarcity and provide comprehensive assessment for embodied AI by offering a large-scale simulation environment with diverse scenes, LLM-driven NPCs, and standardized benchmarks.

How It Works

GRUtopia leverages NVIDIA Omniverse Isaac Sim for physically accurate simulation. Its core innovation lies in the GRScenes dataset, featuring 100k interactive, annotated indoor scenes across 89 categories, enabling broad robot deployment. The GRResidents system integrates LLM-driven NPCs for complex social interactions and task generation, simulating realistic social scenarios. GRBench provides standardized benchmarks for evaluating object navigation, social navigation, and manipulation tasks.

Quick Start & Requirements

• Installation: Follow the provided installation guide locally or via Docker.
• Prerequisites: Ubuntu 20.04/22.04, NVIDIA GPU (RTX 2070+), NVIDIA Driver (535.216.01+ recommended), Python 3.10.16, NVIDIA Omniverse Isaac Sim 4.2.0. Docker and NVIDIA Container Toolkit are optional.
• Assets: Download required assets (~80GB total, ~500MB minimum) using python -m grutopia.download_assets. A user agreement is required for GRScenes-100 access.
• Documentation: https://grutopia.github.io

Highlighted Details

• GRScenes-100: A dataset of 100k interactive, finely annotated indoor scenes.
• GRResidents: LLM-driven NPCs for social interaction and task simulation.
• GRBench: Benchmarks for object loco-navigation, social loco-navigation, and loco-manipulation.
• Teleoperation tools available with Mocap and Apple VisionPro.

Maintenance & Community

The project has released its paper and demos, with a 2.0 release including Gym compatibility and a Pythonic config system. Support is available via a WeChat group.

Licensing & Compatibility

The simulation platform is MIT licensed. The GRScenes dataset is licensed under CC BY-NC-SA 4.0, restricting commercial use.

Limitations & Caveats

The GRScenes dataset is restricted to non-commercial use. Features like vectorized environments, batch execution, and a training framework are listed as future TODOs.