AI-Powered Robotic Arm Control via OpenClaw
Say what you want the arm to do. ClawArm makes it happen.

clawarm.ai · GitHub · Architecture · Safety Guide

---

What is ClawArm?

ClawArm bridges natural language and physical robotic arm motion. Built on top of OpenClaw, it lets you control NERO (7-DOF) and Piper (6-DOF) robotic arms by simply describing what you want — no manual joint calculations, no coordinate math.

Key Features

Feature	Description
Natural Language Control	"Pick up the red block" becomes actual robot motion
Two Control Modes	Skill mode (code generation) + Plugin mode (real-time tools)
Safety Layer	Joint limits, workspace bounds, velocity caps — every command is validated
Mock Mode	Develop and test without physical hardware
Multi-Arm Support	NERO 7-DOF and Piper 6-DOF out of the box
OpenClaw Integration	Works with OpenClaw's web UI, Feishu, Telegram, and more

---

How It Works

"Move the arm to pick up the red block"
         │
         ▼
   ┌─────────────┐
   │  OpenClaw    │  AI agent interprets intent
   │  Gateway     │
   └──────┬──────┘
          │
    ┌─────┴──────┐
    │            │
    ▼            ▼
 Skill Mode   Plugin Mode
 (codegen)    (real-time)
    │            │
    ▼            ▼
 Python       Bridge API
 Script       :8420
    │            │
    └─────┬──────┘
          ▼
   ┌─────────────┐
   │  Safety      │  Joint limits, workspace bounds
   │  Layer       │
   └──────┬──────┘
          ▼
   ┌─────────────┐
   │  Arm Driver  │  CAN bus communication
   └──────┬──────┘
          ▼
      Robot Arm

• Skill Mode: OpenClaw reads the agx-arm-codegen skill, generates a complete Python control script, and executes it. Best for complex, multi-step sequences.
• Plugin Mode: The ClawArm plugin provides arm_move, arm_status, and arm_stop tools that call the bridge server in real time. Best for interactive step-by-step control.

---

Quick Start

Prerequisites

• Linux with CAN interface (SocketCAN)
• Python 3.10+
• OpenClaw installed
• Robotic arm connected via CAN bus

1. Clone and install

git clone https://github.com/Clawland-AI/clawarm.git
cd clawarm
pip install -e ".[dev,arm]"

2. Activate CAN

sudo bash setup/activate_can.sh

3. Start the bridge server

# With real hardware
clawarm-bridge

# Without hardware (mock mode for development)
CLAWARM_MOCK=true clawarm-bridge

4. Install the OpenClaw skill

# Copy skill to OpenClaw workspace
cp -r skills/agx-arm-codegen ~/.openclaw/skills/

# Or install the plugin for real-time tool access
cd plugin && npm install
openclaw plugins install ./plugin

5. Talk to the arm

Open the OpenClaw web UI or send a message via your configured channel:

> Move joint 1 to 0.5 radians, then return to zero position
> Pick up the object at position [0.3, 0.1, 0.2] and place it at [0.3, -0.1, 0.2]
> Draw a small circle in the XY plane at height 0.3m

---

Project Structure

clawarm/
├── skills/                   OpenClaw Skills (Skill Mode)
│   └── agx-arm-codegen/      Natural-language → Python code generation
├── plugin/                   OpenClaw Plugin (Plugin Mode)
│   └── src/tools/            arm_connect, arm_status, arm_move, arm_stop
├── bridge/                   Python Bridge Server (FastAPI)
│   ├── drivers/              Real driver + mock driver for dev
│   └── safety.py             Joint limits, workspace bounds, velocity caps
├── examples/                 Demo scripts
├── workspace/                OpenClaw workspace templates (AGENTS.md, SOUL.md)
├── config/                   OpenClaw configuration templates
├── setup/                    Installation and CAN activation scripts
├── docs/                     Architecture, quickstart, safety docs
└── tests/                    Bridge server and safety tests

---

Supported Arms

Arm	DOF	Joint Control	Cartesian Control
NERO	7	move_j([j1..j7]) radians	move_p/l([x,y,z,r,p,y]) meters/radians
Piper	6	move_j([j1..j6]) radians	move_p/l([x,y,z,r,p,y]) meters/radians

---

Configuration

cp config/openclaw.example.json ~/.openclaw/openclaw.json

Key settings:

{
  "plugins": {
    "entries": {
      "clawarm": {
        "enabled": true,
        "config": {
          "bridgeUrl": "http://localhost:8420",
          "defaultRobot": "nero",
          "safetyEnabled": true
        }
      }
    }
  }
}

---

Safety

ClawArm controls real hardware. Read SECURITY.md before use.

• Safety layer validates all commands before they reach the arm
• Per-robot-type joint angle limits enforced automatically
• Configurable Cartesian workspace boundary box
• Velocity capped at configurable percentage (default 80%)
• Emergency stop via API, OpenClaw tool, or physical button

---

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests (uses mock driver, no hardware needed)
pytest

# Lint
ruff check .

Docker

# Mock mode (no hardware)
docker compose up bridge-mock

# Real hardware (requires --privileged for CAN access)
docker compose up bridge

---

License

MIT — see LICENSE.

clawarm.ai · Built by Clawland AI · Powered by OpenClaw