agent-ros-bridge

Agent ROS Bridge 🔒

The Safety-First Production Gateway for AI-to-Robot Integration
When robots matter, safety comes first.

Why Us? • Quick Start • Safety First • Comparison • Docs

---

Why Agent ROS Bridge?

Unlike diagnostic tools (NASA ROSA) or research platforms (ROS-LLM), Agent ROS Bridge is the only production-ready gateway with built-in safety validation.

The Safety Problem

Deploying LLM-controlled robots in production is dangerous:

• AI hallucinations can damage equipment
• Wrong commands can injure humans
• No validation of AI decisions
• No learning from operator corrections

Our Solution

🛡️ Safety-First Architecture:
├── Human-in-the-Loop (enforced by default)
├── Shadow Mode Validation (200+ hours required)
├── Simulation Testing (10K scenarios, 95.93% success)
├── Gradual Rollout (0% → 100% autonomy)
└── Emergency Stop (always available)

Comparison

Full Comparison →

---

Quick Start

Installation

pip install agent-ros-bridge

Basic Usage

from agent_ros_bridge import RobotAgent

# Create agent with SAFETY enforced
agent = RobotAgent(
    device_id='bot1',
    llm_provider='moonshot',
    require_confirmation=True,  # Human approval required
)

# AI proposes, human approves, robot executes
result = agent.execute("Go to the kitchen")

print(f"Success: {result.success}")
print(f"AI confidence: {result.ai_confidence:.2f}")
print(f"Human approvals: {result.human_approvals}")

Output:

============================================================
🛡️  SAFETY STATUS
============================================================
Device: bot1 (mobile_robot)
Autonomous Mode: False ✅
Human-in-the-Loop: True ✅
Shadow Mode: True ✅
Validation Status: simulation_only
============================================================

🤖 AI Proposal: navigate_to(kitchen)
👤 Human: Approve? (y/n): y
✅ Executed successfully

---

Safety First

Safe-by-Default Configuration

# config/global_config.yaml
safety:
  autonomous_mode: false              # Human approval required
  human_in_the_loop: true             # All AI proposals need approval
  shadow_mode_enabled: true           # Collect validation data
  min_confidence_for_auto: 0.95       # High confidence threshold
  gradual_rollout_stage: 0            # Start at 0% autonomy
  safety_validation_status: "simulation_only"
  required_shadow_hours: 200.0        # Target for validation
  min_agreement_rate: 0.95            # Required agreement %

Deployment Stages

Stage 0: Simulation-Only (Current)
├── ✅ 10K scenarios tested
├── ✅ 95.93% success rate
└── ⚠️ No real-world validation yet

Stage 1: Supervised Operation
├── Human approves all actions
├── Shadow mode collects data
└── Target: 200+ hours, >95% agreement

Stage 2: Gradual Rollout
├── 10% → 25% → 50% → 75% → 100%
├── High confidence only
└── Monitor at each stage

Stage 3: Full Autonomy
├── After validation complete
├── Emergency stop always available
└── Continuous monitoring

Safety Documentation →

---

Architecture

AI Agents ─┬─ WebSocket ─┐
           ├─ gRPC ──────┼──► ┌─────────────┐
           ├─ MQTT ──────┤    │   Gateway   │
           └─ TCP ───────┘    └──────┬──────┘
                                     │
                    ┌────────────────┼────────────────┐
                    ▼                ▼                ▼
            ┌──────────────┐  ┌──────────────┐  ┌──────────┐
            │Safety Layer  │  │ Shadow Mode  │  │  ROS1/   │
            │(Validation)  │  │ (Data Coll.) │  │  ROS2    │
            └──────────────┘  └──────────────┘  └──────────┘

Key Components

---

Features

🛡️ Safety & Validation

• Shadow Mode - Log AI proposals vs human decisions
• Human-in-the-Loop - Enforced by default
• Simulation Testing - 10K scenarios before deployment
• Gradual Rollout - Increase autonomy slowly
• Emergency Stop - Always available

🌐 Multi-Protocol

• WebSocket - Real-time bidirectional
• gRPC - High-performance RPC
• MQTT - IoT messaging
• TCP - Raw socket support

🤖 Universal Support

• Mobile Robots - Navigation, mapping
• Drones - Flight control, aerial missions
• Robot Arms - Manipulation, grasping
• Humanoids - Walking, balancing
• Sensors - Data collection, monitoring

🧠 AI Integration

• Multi-LLM - OpenAI, Moonshot, Anthropic
• Intent Parsing - Natural language understanding
• Context Awareness - Scene understanding
• Multi-Language - English, Chinese, +4 more

---

Documentation

• Safety Guide - Deployment safety requirements
• Architecture - System design
• API Reference - Complete API docs
• Comparison - vs NASA ROSA vs ROS-LLM
• Changelog - Version history

---

Web Dashboard

Agent ROS Bridge includes a modern web dashboard for robot control and monitoring.

Features

• 🎮 Robot Control - D-pad interface, natural language commands
• 📊 Real-time Telemetry - Position, velocity, battery, sensors
• 🧠 Shadow Mode Metrics - AI-human agreement tracking
• 🛡️ Safety Status - Validation gates, deployment readiness
• 🚁 Fleet Management - Multi-robot coordination

Quick Start

Option 1: Docker (Recommended)

# Start bridge + web dashboard
docker-compose --profile web up -d

# Access dashboard
open http://localhost:8081

Option 2: Host (Development)

# Start bridge
agent-ros-bridge --websocket-port 8765

# In another terminal, serve dashboard
cd agent_ros_bridge/web
python3 -m http.server 8081

# Access dashboard
open http://localhost:8081

Dashboard URLs

WebSocket Connection

The dashboard connects to the bridge via WebSocket:

• URL: ws://localhost:8765
• Protocol: JSON messages
• Auth: JWT token (if enabled)

Full Dashboard Guide →

---

Installation

Prerequisites

• Python 3.11+
• ROS1 (Noetic) or ROS2 (Humble/Jazzy)
• Docker (optional, for simulation)

PyPI Install

pip install agent-ros-bridge

Docker Install

docker pull agent-ros-bridge:jazzy-with-nav2
docker run -it agent-ros-bridge:jazzy-with-nav2

Development Install

git clone https://github.com/webthree549-bot/agent-ros-bridge.git
cd agent-ros-bridge
pip install -e ".[dev]"

---

Testing

# Run all tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=agent_ros_bridge --cov-report=html

# Run safety-critical tests only
pytest tests/unit/safety/ -v

# Run simulation tests (requires Docker)
pytest tests/e2e/ -v

Current Status:

• 2,021 tests passing
• 65% code coverage
• Gate 2 validation: PASSED (95.93% success)

---

Safety Certification Roadmap

---

Community

• ROS Discourse: Announcement Thread
• GitHub Issues: Report Issues
• Documentation: Full Docs

---

Citation

If you use Agent ROS Bridge in research, please cite:

@software{agent_ros_bridge,
  title = {Agent ROS Bridge: Safety-First Production Gateway for AI-to-Robot Integration},
  author = {Agent ROS Bridge Contributors},
  year = {2026},
  url = {https://github.com/webthree549-bot/agent-ros-bridge}
}

---

License

MIT License - See LICENSE for details.

---

Built with safety in mind for production robotics deployments.
_{When robots matter, safety comes first.}