This looks good as far as I can tell.
I added the reset of the obstacles to not mantain when a agent hitted an obstacle in the next iteration.
For the rewards, it probably makes more sense to use OnActorEndOverlap? I guess having the Reset is good too to make sure its cleared out.
I have a similar setup to you internally which I haven’t had time to write a tutorial about and found that it worked well with raycasting but not with the array observation. I’m not sure if this is due to a bug in the array observation (although I independently use that observation in a different demo and it worked well so I don’t believe its bugged). So I think it might be a representation issue in the observations or a training issue! Not sure yet.
Let me know what works for you though when you figure it out!
Brendan