How openrobotlab/grutopia works
Project Type and Purpose
• GRUtopia is a full-stack embodied-AI research platform and simulation framework.
• It provides:
• A Python core library (“grutopia/core”) that wraps NVIDIA Omniverse Isaac Sim.
• A modular config system for sensors, robots, scenes, tasks, and metrics.
• A Gym-compatible environment interface (“grutopia/core/gym_env.py”).
• A DataHub for loading assets (GRScenes) and models.
• Runtime managers for single-node or distributed task execution.
• Extension packages (“grutopia_extension”) for LLM-driven NPCs, specialized agents, controllers, benchmarks, and teleoperation servers.
• Toolkits for scene generation and mocap utilities.
File Structure
• Top‐level:
• Dockerfile / CI / pre-commit / setup scripts
• README / docs/ — contains tutorials, API guides, architecture overview, and benchmarks
• grutopia/: core simulation and API
• grutopia_extension/: plug-in modules—agents, controllers, sensors, tasks, interactions
• toolkits/: auxiliary scripts for scene generation, data processing, and mocap
• Key patterns:
• Registry/factory (“grutopia/core/register”) for extensibility
• Clear separation: config / datahub / scene / robot / task / runtime / util
• Gym wrapper around the “core” for RL interoperability
• Demo scripts under grutopia/demo that drive the system end to end
Insights from the README
• Purpose: general-purpose research platform for embodied AGI
• Features: GRScenes dataset, GRResidents (LLM-based NPCs), GRBench for benchmarks
• Dependencies: Isaac Sim, rsl_rl, various vision & LLM libraries, OmniGibson patterns