Reading note
Universal Agent divides control into an environment-dependent module for reaching a target state and a task-dependent module for choosing the next target. The simulated experiments test whether this division allows knowledge about movement to be reused when the task changes.