AI agents trained in simulations that differ from the environments where they are deployed sometimes perform better than agents trained and deployed in the same environment, research shows.
New training approach could help AI agents perform better in uncertain conditions
