Hi everyone, Will using multiple training areas, each with different environment for more dynamic situations be possible? Or the only way is to have multiple training areas that are exactly the same and only change all at once with curriculum learning? Also, the agents are all using camera sensor for visual observations. Multiple training areas should not affect training visual observations in any way right? Thanks!
Having multiple training areas is a way to collect more data. You can train your agents on a distribution of environments but you probably want all of these variations to be very similar. I would recommend your areas to sample their properties from a distribution and sample new properties every time the agent resets rather than having different fixed areas.