ros-bridge（ROS bridge 搜索）

🤖 ROS Bridge Skill

Complete robot-to-agent pipeline. Connect any ROS robot to AI agents with full control.

What's Included

Quick Start

1. Connect to Robot

import { RosbridgeClient } from './rosbridge-client';

const robot = new RosbridgeClient('192.168.1.100', 9090);
await robot.connect();

2. Basic Control

// Move forward 0.5 m/s for 2 seconds
robot.publish('/cmd_vel', 'geometry_msgs/Twist', {
  linear: { x: 0.5, y: 0, z: 0 },
  angular: { x: 0, y: 0, z: 0 }
});

// Turn left at 0.3 rad/s
robot.publish('/cmd_vel', 'geometry_msgs/Twist', {
  linear: { x: 0, y: 0, z: 0 },
  angular: { x: 0, y: 0, z: 0.3 }
});

// Stop
robot.publish('/cmd_vel', 'geometry_msgs/Twist', {
  linear: { x: 0, y: 0, z: 0 },
  angular: { x: 0, y: 0, z: 0 }
});

3. Read Sensors

// Subscribe to LIDAR
robot.subscribe('/scan', 'sensor_msgs/LaserScan', (data) => {
  const minDistance = Math.min(...data.ranges.filter(r => r > 0));
  console.log(`Closest obstacle: ${minDistance.toFixed(2)}m`);
});

// Subscribe to odometry
robot.subscribe('/odom', 'nav_msgs/Odometry', (data) => {
  console.log(`Position: x=${data.pose.pose.position.x}, y=${data.pose.pose.position.y}`);
});

// Subscribe to IMU
robot.subscribe('/imu/data', 'sensor_msgs/Imu', (data) => {
  console.log(`Orientation: ${JSON.stringify(data.orientation)}`);
});

VoltAgent Integration

Turn ROS tools into LLM-callable functions:

import { Agent } from '@voltagent/core';
import { rosTools, initializeROS } from './ros-tools';
import { visionTools } from './tools/vision';
import { sensorTools } from './tools/sensors';
import { explorationTools } from './tools/exploration';

// Connect to robot
initializeROS('ws://192.168.1.100:9090');

// Create embodied agent
const agent = new Agent({
  name: 'ugv-rover-01',
  instructions: `You control a physical UGV robot. You can move, turn, stop,
    read sensors, capture images, and explore autonomously.`,
  llm: anthropic('claude-sonnet-4-20250514'),
  tools: [
    ...rosTools,      // move_forward, turn, stop, read_lidar
    ...visionTools,   // capture_image, detect_objects, detect_faces
    ...sensorTools,   // get_multi_sensor_scan, detect_obstacles_360
    ...explorationTools, // update_map, detect_frontiers, find_safe_path
  ],
});

// Chat with robot
const response = await agent.chat('Move forward slowly and tell me what you see');

Available Tools (20+)

Movement:

• move_forward(speed, duration) — Move at m/s for seconds
• turn(angular_velocity, duration) — Rotate at rad/s
• stop() — Emergency stop

Sensors:

• read_lidar() — 8-direction obstacle scan
• get_multi_sensor_scan() — Combined LIDAR + camera + IMU
• detect_obstacles_360() — Full danger/safe zone map
• estimate_robot_pose() — Position + orientation + velocity

Vision (9 CV Modes):

• capture_image() — Raw camera frame
• detect_objects() — YOLO object detection
• detect_faces() — Face detection
• detect_qr_codes() — QR/barcode scanning
• detect_color(target_color) — Color blob tracking
• detect_lines() — Line following
• get_depth_map() — Depth estimation

Exploration:

• update_map() — Build occupancy grid
• detect_frontiers() — Find unexplored boundaries
• find_safe_path(goal) — A* path planning
• goto(x, y) — Navigate to coordinate

Hardware:

• set_led(color) — Control status LEDs
• set_display(lines) — Write to OLED
• set_gimbal(pan, tilt) — Control camera angle
• play_sound(name) — Audio feedback

Swarm:

• broadcast_position() — Share pose with swarm
• request_assistance(task) — Ask nearby robots for help
• coordinate_formation(formation) — Move in formation

A2A Agent (Discoverable Robot)

Expose your robot as an A2A agent other agents can find and use:

import { createRobotAgent } from './robot-agent';

const agent = createRobotAgent({
  name: 'warehouse-bot-01',
  rosUrl: 'ws://192.168.1.100:9090',
  port: 3001,
  capabilities: ['move', 'scan', 'capture', 'explore'],
});

await agent.start();

// Now discoverable at:
// http://192.168.1.100:3001/.well-known/agent.json

Other agents can now:

import { A2AClient } from 'nanda-ts';

const robot = new A2AClient('http://192.168.1.100:3001');
await robot.sendTask({ message: 'Move to the loading dock' });

X402 Paid Robot Services

Charge other agents for robot API calls:

import { createX402RobotServer } from './x402-wrapper';

const server = createX402RobotServer({
  rosUrl: 'ws://192.168.1.100:9090',
  port: 3002,
  escrowAddress: '0x...',
  pricing: {
    move: 0.01,        // 1¢ per movement
    scan: 0.005,       // 0.5¢ per scan
    capture: 0.02,     // 2¢ per image
    explore: 0.10,     // 10¢ per exploration task
  }
});

await server.start();

Buyers automatically pay via HTTP 402:

const scan = await x402Client.fetch('http://robot:3002/api/scan');
// Paid 0.005 USDC, received LIDAR data

Supported Robots

Works with any ROS1/ROS2 robot running rosbridge:

Prerequisites

1. Robot with ROS — ROS1 Noetic or ROS2 Humble+
2. Rosbridge — rosbridge_websocket running
3. Network access — Agent can reach ws://ROBOT_IP:9090

# On robot (ROS2)
ros2 launch rosbridge_server rosbridge_websocket_launch.xml

# On robot (ROS1)
roslaunch rosbridge_server rosbridge_websocket.launch

Example: Autonomous Exploration

const agent = new Agent({
  name: 'explorer-01',
  instructions: `Explore unknown areas. Use LIDAR to avoid obstacles.
    Build a map as you go. Report interesting findings.`,
  tools: [...rosTools, ...sensorTools, ...explorationTools],
});

// Start autonomous exploration
await agent.chat('Explore this room and map it. Avoid obstacles.');

// Agent will:
// 1. Read LIDAR to detect obstacles
// 2. Find frontiers (unexplored areas)
// 3. Plan path to nearest frontier
// 4. Navigate while avoiding obstacles
// 5. Update map with new data
// 6. Repeat until room is mapped

Links

• rosbridge_suite
• VoltAgent
• QUSD Hardware Skills
• X402 Payments