📷 AnyCam2Ros

Turn Any Camera into ROS2 Image Topics — Unified Pipeline for Any Hardware

English | 中文文档

---

📖 Overview

🎯 What Problem Does This Solve?

When deploying VLA models (like π₀ (pi-zero), OpenVLA) on real robots, or collecting SFT demonstration data for robot learning, you need camera feeds as ROS2 image topics.

But here's the reality:

The Problem:
┌─────────────────────────────────────────────────────────────────────────┐
│  🔄 "I want to align with an existing dataset collected on different    │
│      hardware — how do I replicate the same camera setup?"              │
│                                                                         │
│  🎥 "My data was collected with Insta360 GO 3S, RealSense, USB webcams  │
│      on different machines — I need a unified way to configure them"   │
│                                                                         │
│  ⏰ "Writing cam2image launch files for each camera is tedious"         │
│                                                                         │
│  🔀 "Camera device IDs keep changing after every reboot!"               │
└─────────────────────────────────────────────────────────────────────────┘

AnyCam2Ros provides a unified solution:

The Solution:
┌─────────────────────────────────────────────────────────────────────────┐
│  🎥 Insta360 GO 3S    ─┐                                                │
│  📷 USB Webcam        ─┼──▶  /dev/video*  ──▶  AnyCam2Ros  ──▶  ROS2   │
│  🤖 Any V4L2 Device   ─┘                         CLI           Topics  │
│                                                                         │
│  ✅ Unified config across different hardware                            │
│  ✅ Stable device paths (no more reordering after reboot)              │
│  ✅ One command to configure everything                                 │
│  ✅ Shareable JSON config for dataset alignment                         │
└─────────────────────────────────────────────────────────────────────────┘

AnyCam2Ros = Any Camera → ROS2 Image Topics → VLA Training / Robot Deployment

🤖 Use Cases

Scenario	How AnyCam2Ros Helps
Dataset Alignment	Replicate camera setups from existing datasets on your hardware
VLA Model Deployment	Quickly configure cameras for π₀, OpenVLA, RT-2 deployment
SFT Data Collection	Unified pipeline for collecting manipulation demos
Multi-Camera Setup	Configure 2-4 cameras in minutes with consistent naming
Cross-Machine Sharing	Export/import JSON configs between different robots

---

💡 Why "Any" Camera?

In Linux, everything is a file. If your device can produce video, it becomes /dev/video*.

Device Type	Example	Works with AnyCam2Ros?
Action Camera	Insta360 GO 3S, GoPro (as webcam)	✅ Yes
Depth Camera	RealSense (RGB stream)	✅ Yes
USB Webcam	Logitech C920, generic UVC	✅ Yes
Industrial Camera	FLIR, Basler (with V4L2 driver)	✅ Yes
Phone as Webcam	Android USB Webcam mode, DroidCam	✅ Yes
Capture Card	Elgato, HDMI grabbers	✅ Yes
Virtual Camera	OBS Virtual Cam, v4l2loopback	✅ Yes

If it shows up in /dev/video*, we can publish it to ROS2.

---

✨ Features

Feature	Description
🔍 Auto-Discovery	Scans all /dev/video* devices and shows hardware info
🛡️ Stable Paths	Uses /dev/v4l/by-id so camera order survives reboots
🎨 Beautiful CLI	Rich interactive TUI with tables, spinners, and colors
⚡ Zero Boilerplate	Generates optimized cam2image scripts instantly
📦 Shareable Config	JSON config for team collaboration and dataset alignment

---

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/ly-geming/AnyCam2Ros.git
cd AnyCam2Ros

# Install dependencies
pip install rich

Prerequisites

# Install v4l-utils for camera detection
sudo apt install v4l-utils

# Install ROS2 image_tools
sudo apt install ros-${ROS_DISTRO}-image-tools

Running

python3 scripts/camera_cli.py

The interactive wizard will:

1. Scan — Detect all connected cameras
2. Select — Choose which cameras to configure
3. Configure — Set resolution, FPS, ROS namespace
4. Generate — Create ready-to-run launch scripts

---

📂 Output Structure

generated_cameras/
├── start_cam_front.sh      # Individual camera script
├── start_cam_wrist.sh      # Individual camera script  
└── start_all_cams.sh       # Launch everything with one command

Start all cameras:

./generated_cameras/start_all_cams.sh

Verify with image_view:

ros2 run image_view image_view --ros-args -r image:=/hdas/camera_front/color/image_raw

---

🛠️ Usage Modes

Interactive Mode (Recommended)

python3 scripts/camera_cli.py

Regenerate from Config

Share your cameras.json with teammates or across machines:

python3 scripts/camera_cli.py --from-config

Custom Paths

python3 scripts/camera_cli.py \
  --config /path/to/cameras.json \
  --output-dir /path/to/scripts/

---

📦 Requirements

Dependency	Description
Linux	Required for V4L2 device handling
Python 3.8+	CLI runtime
ROS2	image_tools package
v4l-utils	Camera detection (v4l2-ctl)

---

🤝 Contributing

Contributions are welcome! Feel free to submit a Pull Request.

---

📄 License

MIT © ly-geming

---

⭐ Star this repo if it helps your robot project! ⭐