Add optional ROS 1/2 bridge utilities

[Ikhyeon-Cho]

Hi @koide3,

Based on our discussion in Issue #129, I've implemented the following optional ROS bridge utilities.

Summary

This PR adds header-only conversion utilities between ROS messages (sensor_msgs::PointCloud2/sensor_msgs::msg::PointCloud2) and small_gicp::PointCloud.

• Files:
  • include/small_gicp/ros/ros1.hpp (ROS 1)
  • include/small_gicp/ros/ros2.hpp (ROS 2)
  • include/small_gicp/ros/ros_impl.hpp (Common template implementation)
• Minimal Dependency:
  • No changes to CMakeLists.txt.
  • Users only need to include the header files with standard ROS headers.

Features

• Unified Support: Identical API for both ROS 1 and ROS 2.
• Safety:
  • Automatically filters NaN / Inf points from ROS message
  • Resizes the cloud to fit valid points only
• Performance:
  • Optimized using reserve() + emplace_back() to avoid unnecessary initialization (Tried to align with
    small_gicp's policy of leaving normals/covariances uninitialized until estimation. )

Usage Example

 // ROS 1
 #include <small_gicp/ros/ros1.hpp>
 
 void callback(const sensor_msgs::PointCloud2& msg) {
  
   // XYZ only, NaNs filtered
   auto points = small_gicp::from_ros_msg(msg);
 
   // To estimate normals (required for registration):
   small_gicp::estimate_normals_covariances(*points, *tree);
 
  // Convert back
   auto result_msg = small_gicp::to_ros_msg(*points, "frame_id");
 }

Verification

I verified functional correctness, memory safety, and precision. You can easily test the reproduction code and Dockerfiles in the test/ros-bridge branch of my fork.

Test	Status	Note
ROS 1 Build	✅ Passed	Noetic / g++
ROS 2 Build	✅ Passed	Humble / g++
NaN Filtering	✅ Passed	Invalid points from message filtered out correctly
Round Trip	✅ Passed	Data integrity preserved
Precision	✅ Passed	Float-Double conversion check

[Ikhyeon-Cho]

I ran some performance benchmarks to verify the bridge implementation.

Test Environment

• CPU: AMD Ryzen 9 5900X (12-Core)
• Build: Release (-DCMAKE_BUILD_TYPE=Release)
• Noetic: Ubuntu 20.04 (native)
• Humble: Docker (ros:humble)

Results (~22k points, averaged over 100 samples)

ROS Version	from_ros_msg	to_ros_msg	Total
Noetic	~0.10 ms	~0.04 ms	~0.14 ms
Humble	~0.12 ms	~0.05 ms	~0.15 ms

Notes

• Tested with Velodyne VLP-16 point cloud data
• Sufficient for real-time processing: 10Hz sensor → 100ms budget, bridge uses only ~0.15%
• NaN/Inf filtering included in timing