Skip to content

Latest commit

 

History

History
66 lines (53 loc) · 3.12 KB

requirements.md

File metadata and controls

66 lines (53 loc) · 3.12 KB

System Requirements

Hardware

Component Minimum Recommended
GPU NVIDIA RTX 3000+ (8 GB VRAM) RTX 4070+ (12 GB+ VRAM)
CPU 8-core Intel / AMD 12+ cores
RAM 16 GB 32 GB+
Disk 10 GB SSD 25 GB+ SSD
OS Ubuntu 22.04, macOS 12.6+ Ubuntu 24.04

GPU is optional for basic robot control. Required for perception, VLMs, and AI features.

Tested Configurations

Config GPU CPU RAM Status
Dev workstation RTX 4090 (24 GB) i9-13900K 64 GB ✅ Primary dev
Mid-range RTX 4070 (12 GB) i7-12700 32 GB ✅ Tested
Laptop RTX 4060 Mobile (8 GB) i7-13700H 16 GB ✅ Tested
Headless server No GPU Xeon 32 GB ✅ Control only
Jetson AGX Orin Orin (32 GB shared) ARM A78AE 32 GB ✅ Tested
Jetson Orin Nano Orin (8 GB shared) ARM A78AE 8 GB 🟧 Experimental

Dependency Tiers

Bare pip install dimos installs the core tier. Extras add capabilities on top.

pip install dimos                           # Core only
pip install 'dimos[base,unitree]'             # Full stack + Unitree
pip install 'dimos[base,unitree,sim]'         # + MuJoCo simulation
pip install 'dimos[base,unitree,drone]'       # + Drone support
pip install 'dimos[base,unitree,manipulation]' # + Arm control
Extra What it adds Key packages GPU?
(core) Transport, streams, CLI, blueprints, occupancy maps dimos-lcm, numpy, scipy, opencv, open3d, numba, Pinocchio, typer, textual No
agents LLM agent, speech, tool use langchain, openai, whisper, anthropic No
perception Object detection, VLMs, tracking ultralytics, transformers, moondream Yes
visualization Rerun viewer + bridge rerun-sdk, dimos-viewer No
web FastAPI web interface, audio fastapi, uvicorn, ffmpeg-python No
sim MuJoCo simulation mujoco, playground, pygame No
unitree Unitree Go2 / G1 support unitree-webrtc-connect No
drone DJI Tello / MAVLink drones pymavlink No
manipulation Arm planning + control Drake, piper-sdk, xarm-sdk No
cuda GPU acceleration cupy, onnxruntime-gpu, xformers Yes
cpu CPU inference backends onnxruntime, ctransformers No
misc Extra models, embeddings, hardware SDKs cerebras, edgetam, sentence-transformers, tiktoken Varies
docker Minimal set for Docker sidecar modules dimos-lcm, numpy, opencv-headless, rerun-sdk No
base Kitchen sink (agents + web + perception + viz + sim) All of the above Yes
dev Linting, testing, type stubs ruff, mypy, pytest, pre-commit No
psql PostgreSQL storage psycopg2 No
dds DDS transport (CycloneDDS) dev + cyclonedds No

Headless / Server Environments

If running on a headless Ubuntu server (no display), install OpenGL libraries for visualization dependencies:

sudo apt-get install -y libgl1 libegl1

Nix users (nix develop) don't need this — the flake provides libGL, libGLU, and mesa.