I was thinking of trying a demonstration or imitation approach. This is an easy way to help with exploration, and has helped on M’s Revenge. It would require an offline agent (not IMPALA) of course.
Has anyone experimented with this?
I was thinking of trying a demonstration or imitation approach. This is an easy way to help with exploration, and has helped on M’s Revenge. It would require an offline agent (not IMPALA) of course.
Has anyone experimented with this?