The Embodied Operating System

The open-source unified orchestration layer for Physical AI.
_{意默 — 隐入代码，御机有神}

Documentation · Installation · Quick Start · Recipes · Discord

EMOS is the open-source software layer that transforms quadrupeds, humanoids, and mobile robots into Physical AI Agents. It decouples the robot's body from its mind, providing a hardware-agnostic runtime that lets robots see, think, move, and adapt, orchestrated entirely from simple Python scripts called Recipes.

Write a Recipe once. Deploy it on any robot. No code changes.

from agents.clients.ollama import OllamaClient
from agents.components import VLM
from agents.models import OllamaModel
from agents.ros import Topic, Launcher

text_in  = Topic(name="text0", msg_type="String")
image_in = Topic(name="image_raw", msg_type="Image")
text_out = Topic(name="text1", msg_type="String")

model  = OllamaModel(name="qwen_vl", checkpoint="qwen2.5vl:latest")
client = OllamaClient(model)

vlm = VLM(
    inputs=[text_in, image_in],
    outputs=[text_out],
    model_client=client,
    trigger=text_in,
)

launcher = Launcher()
launcher.add_pkg(components=[vlm])
launcher.bringup()

That's a complete robot agent. It sees through a camera, reasons with a vision-language model, and responds, all running as a managed ROS2 lifecycle node.

EMOS is built on three open-source components that work in tandem:

The three layers form a vertical stack. Sugarcoat provides the execution primitives. Kompass builds navigation nodes on top of them. EmbodiedAgents builds cognitive nodes on the same foundation. At runtime, the Launcher brings everything to life from a single Python script, the Recipe.

Hardware-agnostic Recipes. A security patrol Recipe written for a wheeled AMR runs identically on a quadruped. EMOS handles kinematic translation and action commands beneath the surface.

Event-driven autonomy. Robots must adapt to chaos. EMOS enables behavior switching at runtime based on environmental context, not just internal state. If a cloud API disconnects, the Recipe triggers a fallback to a local model. If the navigation controller gets stuck, an event fires a recovery maneuver. Failure is a control flow state, not a crash.

# Cloud API fails? Switch to local backup automatically.
llm.on_algorithm_fail(action=switch_to_backup, max_retries=3)

# Emergency stop? Restart the planner and back away.
events_actions = {
    event_emergency_stop: [
        ComponentActions.restart(component=planner),
        unblock_action,
    ],
}

GPU-accelerated navigation. Kompass moves heavy geometric planning to the GPU, achieving up to 3,106x speedups over CPU-based approaches. It is the first navigation framework with cross-vendor GPU support via SYCL.

ML models as first-class citizens. Object detection outputs can trigger controller switches. VLMs can alter planning strategy. Vision components can drive target following. The intelligence and navigation layers are deeply integrated through a shared component model.

Auto-generated web UIs. EMOS renders a fully functional web dashboard directly from your Recipe definition, real-time telemetry, video feeds, and configuration controls with zero frontend code.

Install the EMOS CLI from the latest release:

curl -sSL https://raw.githubusercontent.com/automatika-robotics/emos/main/stack/emos-cli/scripts/install.sh | sudo bash

Or build from source (requires Go 1.23+):

git clone https://github.com/automatika-robotics/emos.git
cd emos/stack/emos-cli
make build && sudo make install

Then choose your deployment mode:

# Container mode (no ROS required, runs in Docker)
emos install --mode container

# Native mode (requires existing ROS 2 installation)
emos install --mode native

# Or run without arguments for an interactive menu
emos install

Container mode pulls the public EMOS image from GHCR and runs it in Docker. Sensor drivers must be running externally.

Native mode detects your ROS 2 installation, fetches the EMOS source, installs all dependencies (including GPU-accelerated kompass-core), and builds a workspace at ~/emos/ros_ws.

Install ROS2 and all EMOS dependencies in userspace with no root and no Docker, on any Linux distro:

curl -fsSL https://pixi.sh/install.sh | bash
git clone --recurse-submodules https://github.com/automatika-robotics/emos.git
cd emos && pixi install && pixi run setup

Then enter the environment with pixi shell and source install/setup.sh. See the installation docs for details.

You also need a model serving platform. EMOS supports Ollama, RoboML, vLLM, LeRobot, and any OpenAI-compatible endpoint.

See the CLI documentation for the full command reference.

Recipes are not scripts. They are complete agentic workflows that combine intelligence, navigation, and system orchestration into a single, readable Python file.

The General-Purpose Assistant routes verbal commands to the right capability using semantic intent:

router = SemanticRouter(
    inputs=[query_topic],
    routes=[llm_route, goto_route, vision_route],
    default_route=llm_route,
    config=router_config
)

The Vision-Guided Follower fuses depth and RGB to track a human guide without GPS or SLAM:

controller.inputs(
    vision_detections=detections_topic,
    depth_camera_info=depth_cam_info_topic
)
controller.algorithm = "VisionRGBDFollower"

The documentation includes 16 recipes covering conversational agents, prompt engineering, semantic mapping, tool calling, VLA manipulation, point navigation, vision tracking, multiprocessing, runtime fallbacks, and event-driven cognition.

EMOS publishes an llms.txt, a structured context file designed for AI coding agents. Feed it to your preferred LLM and let it write Recipes for you.

With the EMOS CLI:

emos recipes              # Browse available recipes
emos pull vision_follower  # Download one
emos run vision_follower   # Launch it

Custom recipes go in ~/emos/recipes/<name>/ with a recipe.py and manifest.json. See the CLI guide for details.

EMOS has been developed in collaboration between Automatika Robotics and Inria. Contributions from the community are welcome.

EMOS is a monorepo that integrates three independently developed packages. You can open issues here or directly in the relevant repository:

If you're unsure, just open it on emos and we'll route it to the right place.

Each package has developer documentation with guides for extending the framework:

• EmbodiedAgents developer docs — custom components, clients, models
• Kompass developer docs — algorithms, navigation components, data types
• Sugarcoat developer docs — type system, event internals, testing

MIT. See the LICENSE file for details.