Figure 7.

Cost (negative return) of IMRL agent in the 12 tasks 2D Multi-Valley domain for different intrinsic motivation systems and different lengths of the developmental period. “No Skills” shows the performance of a monolithic agent that does not learn skills and has no developmental period. The horizontal black line shows the average cost of the policy learned by the monolithic agent after 5000 episodes. Shown is the mean over 10 independent runs that have been smoothened by a moving window average with window length 50.