embodied-os（体现操作系统）

Embodied-OS - AI Robot Control System

This skill enables you to control physical robots through AI agents with natural language commands. Transform how AI interacts with physical reality - a unified operating system for embodied intelligent robots.

When to Activate This Skill

Activate this skill when the user:

• Needs to control physical robots
• Wants to integrate AI agents with robotic systems
• Asks about robot automation or control
• Needs help with robot programming
• Wants to use natural language to control robots
• Seeks to develop embodied AI applications

Core Features

✅ Unified Robot Control - Single API for controlling any robot platform ✅ AI Agent Integration - Natural language control like talking to ChatGPT ✅ Multi-Modal Perception - Vision, audio, and tactile sensing ✅ High-Level Actions - Navigation, manipulation, and interaction primitives ✅ Task Planning - AI-powered task decomposition and execution ✅ Safety System - Multi-layer safety guarantees for physical robots

Installation

Step 1: Install the Skill

clawhub install embodied-os

Step 2: Install the Package

Option A: Python (PyPI)

pip install openclaw-embodied-os

Option B: Node.js (npm)

npm install openclaw-embodied-os

Option C: From Source

git clone https://github.com/ZhenRobotics/openclaw-embodied-os.git
cd openclaw-embodied-os
pip install -e .

Step 3: Configure API Keys (Optional for AI agents)

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."

Or create a .env file:

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...

Quick Start

Basic Robot Control

from embodied_os import EmbodiedOS

# Initialize the OS
os = EmbodiedOS()

# Connect to a robot
robot = os.connect_robot(
    platform="simulated",
    model="test_robot"
)

# Control the robot
robot.actions.move_to(x=0.5, y=0.3, z=0.2)

# Detect objects
objects = robot.perception.detect_objects()

# Pick and place
if objects:
    robot.actions.pick(object_id=objects[0].id)
    robot.actions.place(position=(0.7, 0.4, 0.1))

AI Agent Control

from embodied_os import AgentInterface

# Create AI agent
agent = AgentInterface(robot=robot, model="claude-sonnet-4")

# Natural language control
agent.execute("Pick up the red cube and place it in the box")

# Conversation
response = agent.chat("What do you see?")
print(response)

Supported Robot Platforms

Current Support

• Universal Robots (UR3e, UR5e, UR10e)
• Franka Emika Panda
• Boston Dynamics Spot
• Simulated Robots (for testing)

Coming Soon

• ABB robots
• KUKA robots
• Custom robots via plugin system

Use Cases

1. Warehouse Automation

warehouse_robot = os.connect_robot(platform="mobile_manipulator")

agent.execute("""
    Go to aisle 5, shelf B.
    Pick up all items marked with red tags.
    Transport them to the packing station.
    Report the quantity and item IDs.
""")

2. Elderly Care Assistant

care_robot = os.connect_robot(platform="service_robot")

agent.monitor_and_assist("""
    Watch for the person calling for help.
    If they ask for water, bring them a glass.
    If they drop something, pick it up.
""")

3. Research Lab Assistant

lab_robot = os.connect_robot(platform="dual_arm_robot")

agent.execute("""
    Set up the chemistry experiment:
    1. Measure 50ml of solution A
    2. Heat to 60 degrees
    3. Add catalyst
    4. Stir for 2 minutes
""")

Core Capabilities

1. Unified Robot Control Interface

• Single API for controlling any robot
• Works across different platforms and manufacturers
• Plug-and-play integration

2. AI Agent Natural Language Control

• Control robots like talking to ChatGPT
• Supports Claude, GPT, and custom agents
• Context-aware task execution

3. Multi-Modal Perception

• Vision: Camera, object detection, depth sensing
• Audio: Sound capture, speech recognition
• Tactile: Force sensors, contact detection

4. High-Level Action Primitives

# Navigation
robot.actions.navigate_to(x=2.0, y=1.5, theta=0)

# Manipulation
robot.actions.pick(object="cup")
robot.actions.place(location="table")

# Interaction
robot.actions.press_button(target="elevator")
robot.actions.open_door(handle_position=[1.0, 0.5, 1.0])

5. AI-Powered Task Planning

# High-level task
task = "Prepare coffee for the user"

# Automatic decomposition and execution
plan = robot.planner.create_plan(task)
robot.planner.execute(plan, monitor=True)

6. Multi-Layer Safety System

# Define safety constraints
robot.safety.set_workspace_bounds(
    x_min=0, x_max=2.0,
    y_min=-1.0, y_max=1.0,
    z_min=0, z_max=1.5
)

# Force limits
robot.safety.set_max_force(50.0)

# Collision avoidance
robot.safety.enable_collision_avoidance()

Architecture

┌─────────────────────────────────────────────────┐
│              AI Agent Layer                     │
│         (Claude, GPT, Custom Agents)            │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│            Embodied-OS Core                     │
│  ┌─────────┐  ┌─────────┐  ┌────────────┐     │
│  │ Natural │  │  Task   │  │   Safety   │     │
│  │Language │  │ Planner │  │ Validator  │     │
│  └─────────┘  └─────────┘  └────────────┘     │
│  ┌─────────┐  ┌─────────┐  ┌────────────┐     │
│  │Perception│  │ Action  │  │   State    │     │
│  │ Module  │  │Executor │  │  Manager   │     │
│  └─────────┘  └─────────┘  └────────────┘     │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│      Robot Abstraction Layer (RAL)              │
│   Unified interface for all robot types         │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│           Physical Robots                       │
│  Manipulators | Mobile | Humanoids | Drones    │
└─────────────────────────────────────────────────┘

Configuration

Create a config.yaml file:

robot:
  platform: universal_robot
  model: UR5e
  endpoint: 192.168.1.100

perception:
  cameras:
    - name: head_camera
      type: realsense_d435
      resolution: [1280, 720]
      fps: 30

safety:
  workspace:
    x: [0, 2.0]
    y: [-1.0, 1.0]
    z: [0, 1.5]
  max_velocity: 0.5  # m/s
  max_force: 50  # N

agent:
  model: claude-sonnet-4
  api_key: ${ANTHROPIC_API_KEY}

API Reference

Core Classes

EmbodiedOS

Main interface to the system.

os = EmbodiedOS(config_path="config.yaml")
robot = os.connect_robot(platform, model, endpoint)
os.disconnect_all()

Robot

Represents a connected robot.

robot.actions.move_to(x, y, z)
robot.perception.get_image()
robot.state.get_joint_positions()
robot.safety.emergency_stop()

AgentInterface

AI agent control interface.

agent = AgentInterface(model="claude-4", robot=robot)
agent.execute(task_description)
agent.chat(message)

Requirements

• Python: 3.9 or higher
• Dependencies: numpy>=1.20.0, pyyaml>=6.0
• Optional: ROS2 (for ROS integration)
• Optional: CUDA (for vision processing)
• Optional: Anthropic API key (for Claude agent)
• Optional: OpenAI API key (for GPT agent)

Examples

See the examples/ directory:

• basic_control.py - Basic robot control
• agent_control.py - AI agent interaction

Run examples:

python examples/basic_control.py
python examples/agent_control.py

Documentation

• GitHub: https://github.com/ZhenRobotics/openclaw-embodied-os
• npm: https://www.npmjs.com/package/openclaw-embodied-os
• PyPI: https://pypi.org/project/openclaw-embodied-os/
• README: Complete documentation and guides
• QUICKSTART: 5-minute quick start guide

Roadmap

Phase 1: Core Platform (Current)

• [x] Robot abstraction layer
• [x] Basic perception system
• [x] Action executor
• [x] Safety system
• [x] Agent interface

Phase 2: Advanced Features (Q2 2026)

• [ ] Multi-robot coordination
• [ ] Advanced vision processing
• [ ] Learning from demonstration
• [ ] Cloud deployment

Phase 3: Ecosystem (Q3 2026)

• [ ] Skill marketplace
• [ ] Community plugins
• [ ] Simulation environments
• [ ] Mobile app control

License

MIT License - see LICENSE file for details.

Changelog

v0.1.0 (2026-03-08)

• Initial release
• Unified robot control interface across platforms
• AI Agent natural language control
• Multi-modal perception (vision, audio, tactile)
• High-level action primitives
• AI-powered task planning
• Multi-layer safety system
• Support for Universal Robots, Franka Panda, Boston Dynamics Spot
• Simulated robot support for testing

Links

• GitHub: https://github.com/ZhenRobotics/openclaw-embodied-os
• npm: https://www.npmjs.com/package/openclaw-embodied-os
• PyPI: https://pypi.org/project/openclaw-embodied-os/
• Release: https://github.com/ZhenRobotics/openclaw-embodied-os/releases/tag/v0.1.0

Embodied-OS - Making robots as easy to control as talking to a friend.