Fig. 20
From: Hierarchical intrinsically motivated agent planning behavior with dreaming in grid environments
Comparison of agents with a different intrinsic modulation setting in the exhaustible resource experiment. Baseline—the agent without any intrinsic modulation. Negative-empowerment—the agent with intrinsic modulation, where the ideal four-step empowerment is the intrinsic reward (values are shifted to \([-1, 0]\)). Positive-empowerment—the same, but the intrinsic reward is shifted to [0, 1]. Zero-const—the same, but the intrinsic reward equals zero