Algorithm 3.

Extrinsic phase: one step of utility-based planning

if (time_out OR (NOT sub_goal_active)) then	⊳ Select non-achieved sub-goal
(sub_goal_image, sub_goal_position) ← Scan(goal)
focus_image ← ScanEnvironmentWithSameFocusAsSubGoal(environment)
sub_goal_active ← GoalNotAchievedCheck(sub_goal_image, focus_image)
end if
if (sub_goal_active = TRUE) then
(object_image, object_position) ← Scan(environment)	⊳ Select object
(sub_goal_achievable, action) ← ActionPlanning(predictors, action_list,…
sub_goal_image, sub_goal_position, object_image, object_position)	⊳ Plan action
if (sub_goal_achievable = TRUE) then
object_utility ← ComputeUtility(object_affordance, sub_goal_value)
if (max_utility_estimatation = TRUE) then	⊳ Computing the maximum possible utility
if (object_utility ≥ potential_utility) then
potential_utility ← LeakyAverage(potential_utility, object_utility)	⊳ Increase utility expectation
end if
else	⊳ Acting if high utility is attainable
if (object_utility ≥ potential_utility) then
ExecuteAction(object_image, object_position, action)
end if
potential_utility ← LeakyAverage(potential_utility, 0)	⊳ Decrease utility expectation
end if
sub_goal_active = FALSE
end if
end if