Embodied Agents
Embodied Agents is a fully-loaded ROS2 based framework for creating interactive physical agents that can understand, remember, and act upon contextual information from their environment.
Last verified:
What is Embodied Agents?
Embodied Agents (The Embodied Operating System) is an open-source unified orchestration layer for Physical AI that transforms robots into Physical AI Agents. It provides a hardware-agnostic runtime that lets robots see, think, move, and adapt – all orchestrated from pure Python scripts called Recipes. Write a Recipe once and deploy it on any robot, from wheeled AMRs to humanoids, without rewriting code.
Embodied Agents pricing
Pricing model: Freemium
Fully free and open-source under MIT license. No paid tiers or subscription plans. Downloadable from GitHub and available via apt package manager for ROS distributions. All features including EmbodiedAgents, KOMPASS navigation, and Sugarcoat architecture are included at no cost.
Embodied Agents pros
- Open-source with MIT license
- Hardware-agnostic runtime works across any robot platform
- Pure Python Recipes simplify development
- Write once, deploy on wheeled AMRs to humanoids
- GPU-accelerated navigation up to 3,106x faster than CPU
- Event-driven architecture for runtime adaptivity
- Hot-swap ML models on network failure automatically
- Built-in semantic memory with vector databases
- Automatic web UI generation for every Recipe
- Pre-installed on partner robots from inMotion Robotics and DEEP Robotics
- Native ROS 2 integration
- Supports local and cloud ML models (Ollama, vLLM, lmdeploy)
- VLM support for task decomposition
- VLA support for end-to-end manipulation
- Zero frontend code required for web console
Embodied Agents cons
- Requires ROS 2 humble or newer
- Requires separate model serving platform installation (Ollama, vLLM)
- Steep learning curve for robotics newcomers
- GPU required for full navigation capabilities
- Limited documentation compared to mature ROS packages
- Only 27 GitHub stars indicating early adoption
- Python 99.7% codebase limits performance-critical use cases
- Requires Ubuntu for package manager installation
Frequently asked questions about Embodied Agents
What is EMOS?
EMOS (The Embodied Operating System) is an open-source unified orchestration layer for Physical AI. It transforms robots into Physical AI Agents by providing a hardware-agnostic runtime that lets robots see, think, move, and adapt – all orchestrated from pure Python scripts called Recipes.
What robots is EMOS compatible with?
EMOS is hardware-agnostic and works across any robot platform including wheeled AMRs, differential drive robots, Ackermann vehicles, omnidirectional platforms, quadrupeds, and humanoids. Partner robots from inMotion Robotics and DEEP Robotics come with EMOS pre-installed.
What is a Recipe in EMOS?
A Recipe is a pure Python script that defines complete robot behaviors. It can wire together vision-language models, LLMs, GPU-accelerated navigation, end-to-end manipulation via VLA policies, speech, semantic routing, and spatio-temporal memory. Write one Recipe once and deploy it on any robot without rewriting code.
What are the three core components of EMOS?
EMOS is built on three open-source components: EmbodiedAgents (Intelligence layer – agentic graphs of ML models with semantic memory and event-driven reconfiguration), KOMPASS (Navigation layer – GPU-powered planning and control for real-world mobility), and Sugarcoat (Architecture layer – event-driven system primitives and imperative launch API).
How does EMOS handle runtime failures?
EMOS uses an event-driven architecture with automatic fallback mechanisms. Cloud API drops trigger automatic fallback to edge models. When stuck in a cluttered corridor, navigation controllers can be hot-swapped from sensor events. Agents can reconfigure themselves at runtime based on internal and external events.
Do I need to write frontend code for the web console?
No. Every Recipe auto-generates its own real-time web UI with zero frontend code required. The zero-touch web console runs on the robot for browsing and running recipes.
What ML model serving platforms does EMOS support?
EMOS is agnostic to model serving platforms and currently supports Ollama, RoboML, and any platform or cloud provider with an OpenAI compatible API including vLLM and lmdeploy. Support for new platforms is continuously being added.
How do I install EMOS?
For ROS versions >= humble on Ubuntu, install with apt: sudo apt install ros-$ROS_DISTRO-automatika-embodied-agents. Alternatively, download the deb package from the release page or install from source using git clone and colcon build. Can also deploy via Docker, native ROS2, or userspace with Pixi.
What is the licensing for EMOS?
EMOS is fully open-source under the MIT license. All components including EmbodiedAgents, KOMPASS, and Sugarcoat are MIT licensed with no restrictions on commercial or personal use.
Can EMOS agents adapt to different tasks on the same robot?
Yes. The same quadruped can run a warehouse delivery Recipe in the morning and a security patrol Recipe at night without rebuilding the stack. EMOS enables general-purpose software for general-purpose robots by allowing different Recipes to be deployed on the same hardware.