agent-ros-bridge 0.6.0

Universal interface for AI agents to control ROS robots via natural language

Agent ROS Bridge

Universal ROS1/ROS2 bridge for AI agents to control robots and embodied intelligence systems.

---

What It Does

Agent ROS Bridge sits between your AI agent and your robots. It speaks WebSocket, MQTT, and gRPC on the agent side, and ROS1/ROS2 on the robot side — with JWT auth, agent memory, safety confirmation, and fleet orchestration built in.

┌─────────────────────────────────────────────────────────────┐
│                      AI AGENT LAYER                          │
│   LangChain · AutoGPT · Claude (MCP) · OpenClaw · Custom    │
└──────────────────────────┬──────────────────────────────────┘
                           │  WebSocket / MQTT / gRPC
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                   AGENT ROS BRIDGE                           │
│  ┌──────────────┐  ┌─────────────┐  ┌────────────────────┐  │
│  │  Transports  │  │  Core Bridge│  │  AI Integrations   │  │
│  │  WebSocket   │  │  JWT Auth   │  │  Memory (SQLite)   │  │
│  │  MQTT        │  │  RBAC       │  │  Safety Manager    │  │
│  │  gRPC        │  │  Fleet Mgmt │  │  Tool Discovery    │  │
│  └──────────────┘  └─────────────┘  └────────────────────┘  │
└──────────────────────────┬──────────────────────────────────┘
                           │  rclpy / rospy
              ┌────────────┴────────────┐
              ▼                         ▼
       ROS2 (Jazzy/Humble)         ROS1 (Noetic)

---

Feature Status

Feature	Status	Notes
WebSocket transport	✅ Working	Full duplex, TLS support
MQTT transport	✅ Working	paho v2 compatible
gRPC transport	✅ Working	Bidirectional streaming, TLS/mTLS, JWT auth
JWT authentication	✅ Working	Per-transport, RBAC included
Agent memory (SQLite)	✅ Working	TTL, append, list ops
Agent memory (Redis)	✅ Working	Same API, Redis backend
Safety manager	✅ Working	Emergency stop, policy levels
Fleet orchestration	✅ Working	Task allocation, priority queue
LangChain adapter	✅ Working	ROSBridgeTool, ROSAgent
MCP server (stdio)	✅ Working	Claude Desktop integration
AutoGPT adapter	✅ Working	Command discovery & execution
OpenClaw adapter	✅ Working	ClawHub skill + extension support
Web dashboard	✅ Working	HTTP polling, emergency stop button
Simulated robot	✅ Working	No ROS needed — perfect for testing
ROS2 connector	✅ Working	Dynamic message types, publish/subscribe, discovery
ROS1 connector	✅ Working	Full feature parity with ROS2
Tool auto-discovery	✅ Working	Real-time topic/service/action introspection

---

Quick Start (no ROS required)

The fastest path uses the built-in simulated robot — a full bridge with a virtual TurtleBot that responds to commands over WebSocket, with zero ROS installation needed.

1. Install

pip install agent-ros-bridge

2. Set JWT secret

export JWT_SECRET=$(openssl rand -base64 32)

The bridge will refuse to start without JWT_SECRET. This is intentional.

3. Start the simulated bridge

python examples/quickstart/simulated_robot.py
# WebSocket available at ws://localhost:8766

Or with Docker:

docker-compose -f examples/quickstart/docker-compose.yml up

4. Generate a token and connect

# Generate a JWT token
python scripts/generate_token.py --user myagent --roles operator

# Connect with wscat (npm install -g wscat)
wscat -c "ws://localhost:8766?token=<TOKEN>"

5. Send commands

{"command": {"action": "list_robots"}}
{"command": {"action": "get_topics"}}
{"command": {"action": "move", "parameters": {"direction": "forward", "distance": 1.0}}}
{"command": {"action": "rotate", "parameters": {"angle": 90}}}

---

Python API

Basic bridge

import asyncio, os
from agent_ros_bridge import Bridge
from agent_ros_bridge.gateway_v2.transports.websocket import WebSocketTransport

os.environ["JWT_SECRET"] = "your-secret"   # or set in env before importing

async def main():
    bridge = Bridge()
    bridge.transport_manager.register(WebSocketTransport({"port": 8765}))
    async with bridge.run():
        print("Bridge running on ws://localhost:8765")
        while bridge.running:
            await asyncio.sleep(1)

asyncio.run(main())

With LangChain

from agent_ros_bridge import Bridge

bridge = Bridge()

# Get a LangChain-compatible tool
tool = bridge.get_langchain_tool(actions=["navigate", "move_arm", "get_status"])

# Plug into any LangChain agent
from langchain_core.tools import BaseTool
from langchain.agents import AgentExecutor, create_react_agent

agent = create_react_agent(llm, [tool], prompt)
executor = AgentExecutor(agent=agent, tools=[tool])
result = executor.invoke({"input": "Navigate the robot to zone A"})

With Claude Desktop (MCP)

from agent_ros_bridge import Bridge

bridge = Bridge()
mcp = bridge.get_mcp_server(mode="stdio")
await mcp.start()   # Claude Desktop can now control robots

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "agent-ros-bridge": {
      "command": "python",
      "args": ["-m", "agent_ros_bridge.integrations.mcp_transport"],
      "env": {
        "JWT_SECRET": "your-secret-here"
      }
    }
  }
}

With AutoGPT

from agent_ros_bridge import Bridge

bridge = Bridge()
adapter = bridge.get_autogpt_adapter()

# List available commands
commands = adapter.get_commands()

# Execute a command
result = await adapter.execute_command("ros_navigate", target="zone_a")

With OpenClaw

from agent_ros_bridge import Bridge

bridge = Bridge()
adapter = bridge.get_openclaw_adapter()

# Get ClawHub skill path
skill_path = adapter.get_skill_path()
print(f"Skill location: {skill_path}")
# Package for ClawHub: cd {skill_path} && python scripts/package_skill.py

# Or use extension mode
tools = adapter.get_tools()
result = await adapter.execute_tool("ros2_publish", {
    "topic": "/cmd_vel",
    "message": {"linear": {"x": 0.5}}
})

Install via ClawHub:

# When published to ClawHub
npx clawhub@latest install agent-ros-bridge

# Or manually package and install
cd skills/agent-ros-bridge
python scripts/package_skill.py
# Upload dist/agent-ros-bridge.skill to ClawHub

Natural language commands (via OpenClaw):

• "Move forward 1 meter" → publishes to /cmd_vel
• "Navigate to the kitchen" → sends Nav2 goal
• "What do you see?" → captures camera frame
• "Check the battery" → reads /battery_state
• "Emergency stop" → triggers safety e-stop

Agent memory

bridge = Bridge(config={"memory_backend": "sqlite", "memory_path": "/tmp/bridge.db"})

# Memory is automatically attached to Bridge.memory
await bridge.memory.set("last_position", {"x": 3.0, "y": 1.5}, ttl=3600)
pos = await bridge.memory.get("last_position")

# Append to a history log
await bridge.memory.append("agent:user1:actions", {"action": "navigate", "ts": "..."})

Safety confirmation

from agent_ros_bridge.integrations.safety import SafetyManager, SafetyLevel

safety = SafetyManager()
safety.register_policy("move_arm", SafetyLevel.DANGEROUS, "Arm movement requires confirmation")

# Register a callback to receive confirmation requests (e.g. send to operator dashboard)
def on_confirm_request(request):
    print(f"Operator approval needed: {request.id} — {request.message}")

safety.on_confirmation_request(on_confirm_request)

# Approve via request ID
await safety.confirm(request_id)

# Emergency stop
safety.trigger_emergency_stop("Obstacle detected")

Fleet orchestration

from agent_ros_bridge.fleet import FleetOrchestrator, FleetRobot, RobotCapability, Task

orchestrator = FleetOrchestrator()
await orchestrator.start()

# Register robots
await orchestrator.add_robot(FleetRobot(
    robot_id="tb4_01", name="TurtleBot4 #1",
    capabilities=RobotCapability(can_navigate=True, battery_hours=4.0)
))

# Submit tasks
task_id = await orchestrator.submit_task(Task(
    type="navigate", target_location="zone_b", priority=3
))

metrics = orchestrator.get_metrics()
print(f"Fleet: {metrics.idle_robots} idle, {metrics.active_robots} active")

---

Docker

Development / CI (no ROS)

# Build
docker build -t agent-ros-bridge .

# Run with JWT secret
docker run --rm -e JWT_SECRET=$JWT_SECRET -p 8765:8765 agent-ros-bridge

With ROS2 Jazzy

docker build -f docker/Dockerfile.ros2 -t agent-ros-bridge:ros2 .
docker run --rm \
  -e JWT_SECRET=$JWT_SECRET \
  -e ROS_DOMAIN_ID=0 \
  -p 8765:8765 \
  agent-ros-bridge:ros2

docker-compose (profile-based)

# ROS2 bridge only
JWT_SECRET=$JWT_SECRET docker-compose --profile ros2 up

# ROS1 bridge + roscore
JWT_SECRET=$JWT_SECRET docker-compose --profile ros1 up

# ROS2 with TurtleBot4 simulator
JWT_SECRET=$JWT_SECRET docker-compose --profile ros2 --profile sim up

---

Connecting to a Real Robot (ROS2)

ROS2 publish/subscribe implementation is in progress. The following shows the intended workflow:

# 1. Install with ROS2 support
pip install "agent-ros-bridge[ros2]"

# 2. Source your ROS2 environment
source /opt/ros/jazzy/setup.bash   # or humble

# 3. Set config pointing at your ROS domain
cat > config/gateway.yaml << EOF
name: my_robot_bridge
transports:
  websocket:
    port: 8765
    auth:
      enabled: true
      jwt_secret: ${JWT_SECRET}
connectors:
  ros2:
    enabled: true
    options:
      domain_id: 0
EOF

# 4. Start
agent-ros-bridge --config config/gateway.yaml

---

CLI Reference

# Start with default config (WebSocket on 8765, gRPC on 50051)
agent-ros-bridge

# Start with a config file
agent-ros-bridge --config config/gateway.yaml

# Demo mode — WebSocket + greenhouse plugin, no ROS needed
agent-ros-bridge --demo

# Override ports
agent-ros-bridge --websocket-port 9000 --grpc-port 50052

# Print version
agent-ros-bridge --version

---

Environment Variables

Variable	Default	Description
JWT_SECRET	—	Required. JWT signing secret (min 32 chars). Generate: openssl rand -base64 32
BRIDGE_CONFIG	—	Path to YAML/JSON config file
BRIDGE_LOG_LEVEL	INFO	Log level: DEBUG, INFO, WARNING, ERROR
BRIDGE_WEBSOCKET_PORT	8765	WebSocket listen port
BRIDGE_GRPC_PORT	50051	gRPC listen port

---

Examples

Example	Description	Run
examples/quickstart/	Simulated robot — zero ROS needed	docker-compose up
examples/fleet/	4-robot fleet with task scheduling	./run.sh
examples/auth/	JWT auth, API keys, RBAC	./run.sh
examples/arm/	Robot arm control (UR, xArm, Franka)	./run.sh
examples/actions/	ROS Actions (nav2, manipulation)	./run.sh
examples/mqtt_iot/	MQTT IoT devices + bridge	docker-compose up
examples/metrics/	Prometheus metrics + Grafana	docker-compose up
examples/v0.5.0_integrations/	LangChain, AutoGPT, MCP, OpenClaw, Dashboard	python <example>.py

---

Project Structure

agent_ros_bridge/
│
├── gateway_v2/              # Core bridge
│   ├── core.py              # Bridge class — transport/connector/plugin lifecycle
│   ├── auth.py              # JWT authentication + RBAC
│   ├── config.py            # YAML/JSON/env config loader  (BRIDGE_* env vars)
│   ├── __main__.py          # CLI entry point  (agent-ros-bridge)
│   │
│   ├── transports/
│   │   ├── websocket.py     # WebSocket transport  ✅
│   │   ├── mqtt_transport.py# MQTT transport        ✅
│   │   └── grpc_transport.py# gRPC transport        🔧 in progress
│   │
│   ├── connectors/
│   │   ├── ros2_connector.py# ROS2 (rclpy)          🔧 in progress
│   │   └── ros1_connector.py# ROS1 (rospy)          🔧 in progress
│   │
│   └── plugins/
│       └── greenhouse_plugin.py  # Example plugin    ✅
│
├── integrations/            # AI agent integrations
│   ├── memory.py            # AgentMemory — SQLite/Redis with TTL  ✅
│   ├── safety.py            # SafetyManager — confirmation + e-stop ✅
│   ├── discovery.py         # ToolDiscovery — MCP/OpenAI export    ✅
│   ├── langchain_adapter.py # ROSBridgeTool, ROSAgent               ✅
│   ├── autogpt_adapter.py   # AutoGPT command adapter               ✅
│   ├── openclaw_adapter.py  # OpenClaw/ClawHub integration          ✅
│   ├── mcp_transport.py     # MCP stdio server                      ✅
│   └── dashboard_server.py  # aiohttp web dashboard                 ✅
│
├── fleet/
│   └── orchestrator.py      # FleetOrchestrator — multi-robot tasks ✅
│
├── plugins/
│   └── arm_robot.py         # UR / xArm / Franka arm plugin         🔧
│
├── actions/
│   └── __init__.py          # ROS Action client (ROS2 + simulated)  ✅
│
└── skills/
    └── agent-ros-bridge/    # OpenClaw/ClawHub skill                ✅
        ├── SKILL.md         # Skill definition
        ├── references/      # ROS1/ROS2 guides
        └── scripts/         # Packaging script

---

Security

• JWT_SECRET is always required — the bridge refuses to start without it.
• Authentication is opt-in per transport via config (auth.enabled: true). The simulated quickstart runs without auth for ease of development.
• RBAC roles: admin (all), operator (navigate/publish/subscribe), viewer (read-only).
• TLS: pass tls_cert / tls_key in transport config.
• Generate tokens: python scripts/generate_token.py --user robot_agent --roles operator

---

Development

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

# Run all tests
pytest

# Run specific suites
pytest tests/unit/ -v
pytest tests/integrations/ -v        # SafetyManager, Memory (no ROS needed)
pytest tests/integration/ -v         # requires running bridge

# Lint + format
ruff check .
black --check .

# Build package
python -m build

---

Documentation

Document	Description
docs/ARCHITECTURE_V2.md	Three-layer gateway architecture
docs/API_REFERENCE.md	Full Python API reference
docs/NATIVE_ROS.md	Native Ubuntu/ROS installation
docs/MULTI_ROS.md	Multi-robot / multi-ROS-version setup
docs/DOCKER_VS_NATIVE.md	Deployment trade-offs
docs/DDS_ARCHITECTURE.md	ROS2 DDS / domain isolation
docs/troubleshooting.md	Common issues and solutions
CHANGELOG.md	Release history
CONTRIBUTING.md	How to contribute
REVIEW.md	Static + functional code review

---

Contributing

See CONTRIBUTING.md. Priority areas for contribution:

• ROS2 Connector — ros2_connector.py: complete _cmd_publish(), subscribe(), and _ros_msg_to_dict() using rosidl_runtime_py
• gRPC Transport — extract proto/bridge.proto, generate stubs, register service
• Safety approval endpoint — add a WebSocket command (safety.confirm) so operators can approve dangerous actions from the agent side
• Tool auto-discovery — implement _discover_topics/services/actions() in integrations/discovery.py

---

License

MIT License — © Agent ROS Bridge Contributors