Hey folks,
I wanted to start a discussion about running multiple agent systems. When I refer to a multiple agent system, I mean a system that is controlling multiple autonomous vehicles independently while restricting high level control to a single node/computer. An example that I am working on is generating optimal trajectories for multiple vehicles tracking a target. At certain times, I am generating a new trajectory for one or more vehicles. The catch is that the trajectory generation algorithm takes a few seconds to execute which means that without some form of concurrency, no other trajectories can be planned. This can cause issues in real time simulation and can even cause safety issues in a real system that may need to react to a new obstacle.
Since there is some research moving in the direction of controlling multiple autonomous vehicles, I thought it would be beneficial for us to discuss some possibilities here. Note that I am hoping to discuss high level algorithms such as optimal trajectory generation (that require knowledge of all vehicle’s trajectories) rather than lower level feedback controllers (which could only require immediate feedback of a single vehicle).
Some approaches I believe would work well include:
- ROS - Have ROS handle the parallel tasks by creating each autonomous agent as a ROS node and running them all at the same time. This solution seems quite elegant however it does require that the system have ROS installed and any potential users of the code understand how to work with ROS.
- Multithreading/Multiprocessing - Speaking specifically about Python (but this should generalize to other languages), it is possible to use the multithreading or multiprocessing packages to run the algorithms in parallel. This solution requires that the language being used has some form of parallel library available. It can also be difficult to debug and tough for beginners to understand.
