Expertise and Robot motor learning:

We often take our expertise in complex motor acts for granted, and it is only when we try to teach a robot to do a similar action, that we notice the complexity of the learning exercise. 
<br>
The following report answers some questions about robot motor learning and expertise.<br><br>



1.To me, the toughest steps in instructing the robot to grasp a pen would be the steps involving sensory feedback: grasping the pen(which, without proprioceptive or pressure feedback, would take ages to learn correctly)
<br>
and rotating the pen between effectors, since this also involves relaxing the effectors just enough for the pen to rotate, but not so much that it slips. Such an act could only be done with sensory feedback. 
<br>
Even so, the ease with which we modulate our fine motor movements dynamically in response to the pen would be unequalled.
<br><br>


2.Although in the case of a robot, a distinction between implicit and explicit steps in the execution may not be clearcut, one could say that initial instructions, such as grasp pen with effectors, or move to coordinates (x,y), i.e. steps that the user specifies from the outset may be considered explicit <br>

whereas aspects of the execution that the robot must learn through trial and error, through training or through other corrective methods, such as how much force to apply, when to stop applying it etc. may be implicitly acquired. It's possible that this information is later computed and stored explicitly, but this storage may not be necessary for the robot to execute this aspect, hence it is implicit.<br>

A chunk, to me, is a single access point to a set of stored values/routines, i.e. a chess player may use just 5 access points to access five configurations of 6 pieces each, whereas a normal person uses each chunk to access a single piece. Since the robot needs only to trigger a single operation, using a single access point, I'd say that there is only onw chunk. If however the robot was learning different subparts of the task in different trials, it may have to access more than one chunk simultaneously.<br>
<br>



3.While I believe human learning does follow similar reward based processes, I believe that in the human case it is done so at many levels. There may be intrinsic "motivations" or feedback mechanisms that reinforce small discrete actions that are merely steps in a sequential act, there may be global feedback and feedforward connections that allow a choice between various subroutines and routines, and finally there may also be external feedback i.e. reward or goal orientation.<br>

In the firefighter's case there may be a global fight-or-flight motivation to survive, in terms of the learning process itself, this may be motivating sensory-motor feedback learning that the person gets while combating the flames. Hence the person learns to adapt to the flames' behaviour and eventually gains expertise in this act.