The vision sensor I use for each agent is quite processing intensive, so I figured it would improve the frame rate if the agents didn't all collect their observations on the same environment step, but I haven't found a way to accomplish this. Is it possible? E.g. my agents each have their DecisionRequester set to 5 and I was hoping that if I instantiated 5 agents 1 frame after each other there would be one observation for the CPU to process per frame, but the Academy seem to demand observations to happen in sync for all agents.
Asked
Active
Viewed 121 times
