Technical Paper | Documentation | X

Openmind's OM1 is a modular AI runtime for agents and robots with multimodal capabilities including movement and speech.

OM1 enables the creation and deployment of AI agents across both digital and physical environments. This means you can design a single AI agent and deploy it not only in the cloud but also on a variety of physical robot platforms, including Quadrupeds, with future support for TurtleBot 3 and Humanoids. This flexibility allows for seamless integration and testing of AI capabilities across different domains, from simulation to real-world applications.

An AI agent built on OM1 is capable of ingesting data from multiple sources (the Web, X/Twitter, cameras and LIDAR) and performing various actions like tweeting, exploring your house, shaking your hand or talking to you. With OM1, you can talk to OpenAI's gpt-4o and literally shake hands with it.

Capabilities of OM1

• Modular Architecture: Designed with Python for simplicity and seamless integration.
• Data Input: Easily handles new data.
• Hardware Support via Plugins: Supports new hardware through plugins for API endpoints and specific robot hardware connections to ROS2, Zenoh, and CycloneDDS.
• Web-Based Debugging Display: Monitor the system in action with WebSim (available at http://localhost:8000/) for easy debugging through visuals.
• Pre-configured Endpoints: Supports Voice-to-Speech, OpenAI’s gpt-4o, DeepSeek, and multiple Visual Language Models (VLMs) with pre-configured endpoints for each service.

Architecture Overview

Hands-On with OM1: 'Hello World' program

Let's create an AI agent named Spot. The Spot agent uses your webcam to capture and label objects. These text captions are then sent to OpenAI 4o, a large language model (LLM) which then returns movement, speech and face action commands. These commands are displayed on WebSim along with basic timing and other debugging information.

1. Clone the repo.

git clone https://github.com/OpenmindAGI/OM1.git
cd OM1
git submodule update --init
uv venv

Note 1: You will need the Rust Python package manager uv.

• To install on Mac Intel chip, use brew install uv.
• To install on Mac M1/2/3, use curl -LsSf https://astral.sh/uv/install.sh | sh (make sure the terminal and python are arm64 native)
• On Linux, use curl -LsSf https://astral.sh/uv/install.sh | sh.

Note 2: If your system doesn't have portaudio, you should install it to run the program.

• To install on Mac, use brew install portaudio
• On Linux, use sudo apt-get install portaudio19-dev python-all-dev

Similarily, you may need ffmpeg, and/or other modules, depending on your system.

2. Set the configuration variables.

Locate the config folder and add your Openmind API key in /config/spot.json. If you do not already have one, you can obtain a free access key at https://portal.openmind.org/.

Note: Using the placeholder key openmind-free will generate errors.

# /config/spot.json
...
"api_key": "om1_live_e4252f1cf005af..."
...

3. Run the spot agent.

uv run src/run.py spot

Please wait for a short while before real time information is displayed.

INFO:root:SendThisToROS2: {'move': 'dance'}
INFO:root:SendThisToROS2: {'speak': "Hello, it's so nice to see you! Let's dance together!"}
INFO:root:SendThisToROS2: {'face': 'joy'}
INFO:root:VLM_COCO_Local: You see a person in front of you.
INFO:httpx:HTTP Request: POST https://api.openmind.org/api/core/openai/chat/completions "HTTP/1.1 200 OK"
INFO:root:Inputs and LLM Outputs: {
	'current_action': 'wag tail', 
	'last_speech': "Hello, new friend! I'm so happy to see you!", 
	'current_emotion': 'joy', 
	'system_latency': {
		'fuse_time': 0.2420651912689209, 
		'llm_start': 0.24208617210388184, 
		'processing': 1.4561660289764404, 
		'complete': 1.6982522010803223}, 
	'inputs': [{
		'input_type': 'VLM_COCO_Local', 
		'timestamp': 0.0, 
		'input': 'You see a person in front of you.'}]
	}

4. WebSim

Go to http://localhost:8000 to see real time logs along with the input and output in the terminal. For easy debugging, add --debug to see additional logging information.

Congratulations! - you just got started with OM1 and can now explore its capabilities.

What's Next?

• Try out more examples, such as uv run src/run.py conversation.
• Add new inputs and actions.
• Design custom agents and robots by creating your own json config file with different combinations of inputs and actions, based on your needs.
• Change the system prompts in the configuration files (located in /config/) to create new behaviors.

Interfacing with New Robot Hardware

OM1 assumes that robot hardware provides a high-level SDK that accepts elemental movement and action commands such as backflip, run, gently pick up the red apple, move(0.37, 0, 0), and smile. An example is provided in actions/move_safe/connector/ros2.py:

...
elif output_interface.action == "shake paw":
    if self.sport_client:
        self.sport_client.Hello()
...

If your robot hardware does not yet provide a suitable HAL (hardware abstraction layer), traditional robotics approaches such as RL (reinforcement learning) in concert with suitable simulation environments (Unity, Gazebo), sensors (such as hand mounted ZED depth cameras), and custom VLAs will be needed for you to create one. It is further assumed that your HAL accepts motion trajectories, provides all battery and thermal management/monitoring, and calibrates and tunes sensors such as IMUs, LIDARs, and magnetometers. OM1 can interface with your HAL via USB, serial, ROS2, CycloneDDS, Zenoh, or websockets. For an example of an advanced humanoid HAL, please see Unitree's C++ SDK. Frequently, a HAL, especially ROS2 code, will be dockerized and can then interface with OM1 through DDS middleware or websockets.

Recommended Development Platforms

OM1 is developed on:

• Jetson AGX Orin 64GB (running Ubuntu 22.04 and JetPack 6.1)
• Mac Studio with Apple M2 Ultra with 48 GB unified memory (running macOS Sequoia)
• Mac Mini with Apple M4 Pro with 48 GB unified memory (running macOS Sequoia)
• Generic Linux machines (running Ubuntu 22.04)

OM1 should run on other platforms (such as Windows) and microcontrollers such as the Raspberry Pi 5 16GB.

Detailed Documentation

More detailed documentation can be accessed at docs.openmind.org and in this repo.

Highlights:

• Core Architecture and Runtime Flow
• Quick Start
• Developer Guide
• Example 1 - Using DeepSeek or Gemini as the Core LLM
• Example 2 - Using Cloud Endpoints for Voice Inputs and Text to Speech
• Example 3 - An Interactive Smart Toy
• Robotics - Unitree Dog
• Robotics - Unitree Dog with Wallet

Contributing

To contribute to this project, follow these steps:

1. Fork the repository: Go to the project's GitHub page and click the "Fork" button in the top-right corner. This will create a copy of the project in your own GitHub account.
2. Create a feature branch: In your forked repository, create a new branch for your changes. This branch should be named something like feature/your-feature-name or fix/your-fix-name. This helps to keep your changes organized and makes it easier to manage multiple contributions.
3. Make your changes: Make the necessary changes to the code in your feature branch. Ensure that your changes are well-documented and follow OM1's coding style.
4. Submit a pull request: Once you've made your changes, submit a pull request to the original repository. This will notify the maintainers of your changes and allow them to review and discuss your contribution.

License

This project is licensed under the terms of the MIT License, which is a permissive free software license that allows users to freely use, modify, and distribute the software. The MIT License is a widely used and well-established license that is known for its simplicity and flexibility. By using the MIT License, this project aims to encourage collaboration, modification, and distribution of the software.