In executing classical plans in the real world, small discrepancies be
tween a planner's internal representations and the real world are unav
oidable. These can conspire to cause real-world failures even though t
he planner is sound and, therefore, ''proves'' that a sequence of acti
ons achieves the goal. Permissive planning, a machine learning extensi
on to classical planning, is one response to this difficulty. This pap
er describes the permissive planning approach and presents GRASPER, a
permissive planning robotic system that learns to robustly pick up nov
el objects.