Multi-agent reinforcement learning in BlueSky

[jooste]

Drop by for all your discussions on reinforcement learning applications in BlueSky! Let's start with some links to previous work:

Bart Vonk's thesis work (BlueSky fork)

Dennis van der Hoff thesis work (BlueSky fork)

Marc Brittain's PhD work 1, 2 (BlueSky fork)

Marta Ribeiro's PhD work 1 (BlueSky fork)

[LuisNavarroGarcia]

Hi!

I think this this is the right place to post about it, but if preferred in some other place I will delete it from here.

I've been working on the topic for my bachelor's final year thesis, when trying to visualize trajectories I open a client and I connect it to a running server (The GUI at start-up shows a pop-up window with the available port, that corresponds to the running server and I just click on that). The simulation is being run with an environment by RLlib but the client doesn't seem to display anything related to the simulation so I don't really know if I'm missing something. Any clarification would be of great help as after several tries I've run out of ideas of what I am doing wrong.

I've been looking at Dennis van der Hoff's work and it has been of great help so, thank you very much for sharing!

Regards,

Luis.

[jooste]

Hi Luis,

Welcome to the bluesky discussion board! The problem you describe is difficult to investigate remotely, since there isn't a traceback or other visual error to analyse. If you are certain that you are connecting to the correct bluesky simulation with your client, the first thing I would check is to put some print statements on the sim side and on the gui side to see:

• If the simulation is really running, and if it contains aircraft
• If the simulation is sending data
• If the client is receiving data at all
  Hopefully that reveals something we can fix.

[LuisNavarroGarcia]

Thank you for the response!

¡I will try my best! In case I find something I'll open a new issue but it is most likely my own fault.

Regards,

Luis.