Answer - 1:
The most complicated tasks for a robot to follow are the ones involving grasping and relative motion between the pen and the point of contact (affectors), be it translational or rotational.
Steps 3 and 7: Here the complication lies in applying just the right amount of pressure so as to allow relative motion (rotation) and at the same time preventing from slipping. This is quite tough to achieve even if sensory feedback is available for reinforcement learning.
Step 13: Tracing the locus over the plane of paper and simultaneously maintaining the contact of wrist with surface is extremely difficult for robots as along with right amount of forces it requires precise balancing as well.
Answer - 2:
The learning paradigm followed by the robot in the experiment involves both explicit and implicit aspects of learning. Explicit knowledge can be associated with the instructions given to the robot, determining the locations of pen on the plane, getting its orientation (through coordinates of end) etc. On the other hand, the information achieved through reinforcement learning for instance the magnitude of forces to apply and its variation with time, speed of moving affectors etc. constitute implicit learning. These are learnt through training and may vary with physical conditions during the experiment.
The chunks in the following case may be the distribution of force values over different affectors along with some goodness measure (i.e. the corresponding result produced - success or failure). This wellness measure helps the robot to retrieve the useful chunk among the learnt ones.
Answer - 3:
The arguments that popped up in our discussion involved classical (implicit) as well as instrumental (explicit) learning. The basic driving force in both of these is the 'reward' that we get out of it. Reward may be positive as well as negative, which determines the further involvement in any activity.
However, 'reward' need not be explicit, rather it can be intrinsic as well leading to gratification. It can be in terms of abstract notions like joy, pleasure that one gets out of any task, for instance, pursuing any sports as a hobby.
As far as the learning experience of fire-fighting expert is concerned, he tries to identify and attack the 'rewarding spots' which can yield maximum gain in extinguishing the fire. The reward in this case is the positive feedback (i.e. fire goes down) which he receives over the task. But for the psychologist (as in the experiment), reward can just be the intrinsic satisfaction which he would gain by extinguishing the fire.