Skip to main content
. 2019 Mar 28;6:20. doi: 10.3389/frobt.2019.00020

Figure 17.

Figure 17

A naive policy in the mountain car task applies a full thrust to the right (right column). This naive policy is unable to make the agent traverse the barrier raised by the gravitational force. In contrast, active inference (left column) automatically infers that the agent should first move left (away from its goal) and subsequently use its momentum (and full forward thrust) to traverse the barrier.