Library that merges the ROS and Boost.Asio event loops in a single thread. No threading required, best effort latency on callback execution.
In your main.cpp init the node in the following way :
int main(int _argc, char ** _argv)
{
// Boost Event loop handler
#if BOOST_VERSION <= 106600
// Version 1.65 on ROS Melodic
auto io_context = std::make_shared<boost::asio::io_service>();
#else
// Version 1.71 on ROS Noetic
auto io_context = std::make_shared<boost::asio::io_context>();
#endif
// Static Function Call doing the magic to unify event loops
AsioCallbackQueue::replaceGlobalQueue(io_context);
ros::init(argc, argv, "test_node", ros::init_options::NoSigintHandler);
ros::NodeHandle nh("TestNode");
TestNode node(*io_context, nh);
io_context->run();
ros::shutdown();
return 0;
}
The boost asio library can be useful for ROS nodes using network sockets,timers...
Unfortunatly both boost asio and ROS need to have their own event loop spinning : ros::spin() for ROS, io_service.run() for boost Asio.
Multiple methods can be used to take care of both event loops :
Use the boost event loop and have a boost timer call cyclically ros::spinOnce().
However this method reduces the application reactivity to the frequency of the pollingTimer.
Use one thread for each event loop. Using this method the callbacks from boost Asio and from ROS are called without any latency. The ROS callbacks are called from another thread, so we have to take care to protect any shared resources with locking mechanisms like mutexes or mailboxes.