Q) Which two instructions in the "programming language" of the 2011 HW would be the most difficult for robots to follow?
A) Most of the tasks would be difficult for robots. And more so for ones which do not have any sensory inputs and must thus rely only on Kinematics.
Furthermore, the material of the robotic hand is also essential. The friction coefficiant, its elasticity and such parameters are important. After all one view is that the main purpose of fingerprints is to enhance tactile sensation for discriminating surface textures. [1] Fingerprints in their current form seem to have emerged primarily in primates.
And degrees of freedom is also a matter to consider. Since Kalakrishnan's robot is lacking in all of these faculties it is not hard to see it has trouble picking up a pen!
But all these are essentially techonological issues. Let us assume that these are not as much of a hindrance. Then which steps would be most difficult?
In general, and I reiterate what was discussed in our group plus also in class, I believe the steps which involve relative motion being the affectors and the object are crucial and needs precision surpassing the other tasks (Step 7). Also steps involving pivoting with surface before writing is equally tough. To write while keeping the hand pivoted is a non trivial task since the forces should be balances at pivot without compromising on the degree of freedom needed to write (Step 13). Of course writing is most tough.
Q) The robot following the learning paradigm as in Kalakrishnan is clearly gaining some expertise. Which aspects of the execution may be called implicit or automatic, and which aspects may be more explicit? What could be the "chunks" in this structure?
A) Attributing implicit/explicit to robots isn't very comforting for me. Nonetheless, the explicit part may be considered the algorithm. Go to the pencil, position itself before it. (I did not notice rotation in the robotic arm and thus assume hat it does not try alligning to the axis of the pencil ans thus the claim.) But the weights learned from the training set or trails if you will, which we may also call grounding can be considered implicit. It learns the relative forces to use and also perhaps the order of force to use, just as expertise in humans often is backed by hours of practicing.
Finally chunking is done when their is some relation between the information within the chunk. Here the relative forces applied may be chunked, thus, without which the pencil spins out of the robot's reach. Also another chunk may be simply just the order of magnitude of force used since too much or too less, both make it hard to perform subsequent tasks.
Q) Comment on whether human learning may also be following similar "reward" based processes? Consider the learning process for the fire-fighting expert who knows how to fight complex fires.
A) Specially after listening to Diksha's presentation on Ritter's paper [2] which spoke of Intrinsic motivation, this question was one which entailed into a long discussion. What we thought was that we must not categorise human learning into reward based and non-reward based, and that essentially all process are reward based. We rather tried to think about the nature and changing nature of the reward (Internal/External). How a basketball player first is rewarded when learning the basics when he puts in a basket. Later, in a team, the reward is not so for individual but for team cooperation. These are nonetheless External rewards. But there can be internal rewards which is why we all pursue the hobbies we like. Rewards can be from appreciation, reassurance or simply self-satisfaction (Internal rewards). One must note that in the context of Internal rewards, if we consider Dopamine the physical manifestation of the reward syste,, it only reinforces neurons on a local level and is unaware as much as the neurons themselves which task is being 'rewarded'.
Sometimes even pinpointing the reward may be difficult as in the case of Fire experts. A novice using logic and proceding by considering the reward to be the immediate lessening of fire doesn't go far. It needs 'intuition' which is just another word for training. The fire fighter need to predict the fire's growth to some extent.
What we concluded upon was, this question wasn't as easy to tackle at all. Not only does the reward keep changing but also os may be hard to realise what it is.