🦾 RobotClaw

Let AI Control Real Robots — From Code to Physical World in One Line

---

🧠 One pip install away from giving your AI agent a real body.

RobotClaw is a Python robotics framework that bridges AI agents with physical robots. It turns high-level AI commands into real-world servo movements — enabling your AI to walk, wave, dance, and interact with the physical world.

Built for the OpenClaw AI agent platform. Compatible with LOBOT LX series bus servos. Designed for makers, researchers, and anyone who wants to bring robots to life.

pip install robotclaw

🎬 What Can You Do?

# Your AI agent says: "让机器人挥手"
# RobotClaw makes it happen in the real world:

from robotclaw import Robot, DEFAULT_CONFIG

with Robot(config=DEFAULT_CONFIG, port="COM3") as robot:
    robot.go_home(time_ms=2000)          # Stand up
    robot.set_joint("left_hip_pitch",    # Move a joint
                    position=350, 
                    time_ms=500)

🤖 User:  "让机器人回到初始位置"
🧠 AI:    → robot.go_home(2000) ✅

🤖 User:  "播放走路动作，2倍速"
🧠 AI:    → player.play(walk_clip, speed=2.0) ✅

🤖 User:  "检查所有舵机的电压"
🧠 AI:    → robot.scan() → 报告电压状态 ✅

No robotics PhD required. If you can write Python, you can control a robot.

---

✨ Why RobotClaw?

Feature	Description
🧠 AI-Native	First-class integration with OpenClaw — AI agents control robots via natural language
🔌 Plug & Play	Connect USB, pip install, and you're controlling servos in 3 lines of code
🎬 Motion Teaching	Record human demonstrations, save as JSON, replay at any speed
🤖 High-Level API	Think in joints (left_hip_pitch), not raw servo IDs
🛠️ CLI Tools	robotclaw-scan for diagnostics, robotclaw-teach for interactive teaching
⚡ Thread-Safe	Production-grade mutex-protected serial communication
🐍 Pure Python	Zero compiled dependencies — runs anywhere Python runs

---

🚀 Quick Start

1. Install

pip install robotclaw

2. Connect & Control

from robotclaw import ServoBus

bus = ServoBus()
bus.connect("COM3", baudrate=115200)

# Move servo to position 500 in 1 second
bus.move(servo_id=1, position=500, time_ms=1000)

# Read current position
pos = bus.read_position(servo_id=1)
print(f"Servo position: {pos}")

bus.disconnect()

3. Build a Walking Robot

from robotclaw import Robot, DEFAULT_CONFIG

with Robot(config=DEFAULT_CONFIG, port="COM3") as robot:
    # All 10 joints to home position
    robot.go_home(time_ms=2000)

    # Control individual joints by name
    robot.set_joint("left_hip_pitch", position=350, time_ms=500)
    robot.set_joint("right_knee", position=600, time_ms=500)

    # Read all joint positions at once
    positions = robot.get_positions()
    print(positions)

4. Teach Your Robot New Moves

No programming needed — just physically move the robot and record:

from robotclaw import Robot, DEFAULT_CONFIG
from robotclaw.recorder import MotionRecorder, MotionPlayer

with Robot(config=DEFAULT_CONFIG, port="COM3") as robot:
    # Step 1: Record a motion by posing the robot
    recorder = MotionRecorder(robot)
    recorder.start_recording("wave_hand")

    robot.unload_all()  # Release servos so you can pose by hand
    input("Pose the robot → Press Enter to capture...")
    recorder.capture_frame()

    input("Next pose → Press Enter...")
    recorder.capture_frame()

    clip = recorder.finish_recording()
    recorder.save(clip, "motions/wave_hand.json")

    # Step 2: Replay at any speed
    robot.load_all()
    player = MotionPlayer(robot)
    player.play(clip, speed=1.5)  # 1.5x speed playback

5. CLI Tools

# Discover all connected servos
robotclaw-scan --port COM3

# Interactive motion teaching terminal
robotclaw-teach --port COM3

---

🧠 AI Integration — The Killer Feature

RobotClaw is designed from the ground up for AI-driven robotics. It ships with a built-in OpenClaw skill that lets AI agents understand and control the robot through natural language:

from robotclaw.openclaw_skill import OpenClawSkill

# Register with your AI agent
skill = OpenClawSkill(port="COM3")

# Now your AI can:
# - Move joints by name
# - Play recorded motions
# - Read sensor data (position, voltage, temperature)
# - Perform diagnostic scans
# - Chain complex movement sequences

Use Cases:

• 🎓 Education — Students learn robotics through conversation with AI
• 🏭 Prototyping — Rapidly test robot behaviors via natural language
• 🎮 Entertainment — AI-controlled robot performances and interactions
• 🔬 Research — Quickly iterate on movement patterns without manual coding

---

🔧 Supported Hardware

Component	Specification
Servos	LOBOT LX-16A / LX-224 / LX-225 bus servos
Interface	USB-to-TTL serial adapter
Baudrate	115200 (direct) or 9600 (controller board)
Power	6–8.4V DC, ≥5A recommended for 10 servos
Default Config	10 servos, 5 per leg (biped robot)

💡 Tip: More servo types and robot configurations coming soon. PRs welcome!

---

📁 Architecture

src/robotclaw/
├── __init__.py          # Public API: ServoBus, Robot, DEFAULT_CONFIG
├── servo_bus.py         # LOBOT LX protocol — the low-level driver
├── robot_config.py      # JointConfig, LegConfig, RobotConfig
├── robot.py             # High-level robot controller
├── cli.py               # CLI: robotclaw-scan, robotclaw-teach
├── openclaw_skill.py    # OpenClaw AI agent skill adapter
├── SKILL.md             # OpenClaw skill descriptor
└── recorder/            # Motion teaching subsystem
    ├── motion_data.py   # Keyframe & MotionClip data models
    ├── recorder.py      # Record motions from physical teaching
    └── player.py        # Play back motions with speed control

---

🗺️ Roadmap

• LOBOT LX bus servo protocol driver
• High-level joint-based robot API
• Motion recording & playback system
• OpenClaw AI agent integration
• CLI diagnostic & teaching tools
• 🔜 Visual motion editor (web UI)
• 🔜 Inverse kinematics engine
• 🔜 Support for more servo protocols (Dynamixel, Feetech)
• 🔜 ROS 2 bridge
• 🔜 Reinforcement learning integration

---

🤝 Contributing

We welcome contributions from robotics enthusiasts, AI researchers, and Python developers!

# Clone and install in development mode
git clone https://github.com/RobotBase/robotclaw.git
cd robotclaw
pip install -e ".[dev]"

# Run tests
python -m pytest tests/ -v

See the CHANGELOG for recent updates.

---

📄 License

MIT License — Use it freely in your projects, commercial or otherwise.

---

Built with ❤️ by the RobotBase team
_{Making AI-powered robotics accessible to everyone}