Motor Expertise

-- Pankaj Prateek    

Ans 1. For a robot, the steps involving tactile response would be tougher than the others due to the complexity associated with them. In the given experiment, the steps involving relative motion (rotational or translational) between the point of contact (the fingers and thumb) and the object (the pen) and also grasping the pen would be the toughest ones. While the former involves precise motion of the pen in the 3-D space while applying just enough torque so that the pen doesn't rotate (this requires constant feedback from the environment and conversion of this to motor actions), the latter involves application of exact pressure to hold the pen steady. I also feel that the balancing of the pen on the paper would also be difficult due to the same reasons. (Steps 3,7&13)

Ans 2. Explicit content consists of the instructions given to the bot, like the knowledge of the trajectory, the directions of forces, the location and orientation of pen etc. This gives it an idea of how it has to move its fingers in order to grab the handle or the pen.
The implicit knowledge consists of the machine learning algorithms that help in implicit learning. This includes the correct magnitude of pressure required to hold the pen, the velocity of its "hands" etc. These are learnt through trial and error and vary from bot-to-bot and with the conditions.
Each instruction, with its own set of parameters, is a chunk. These learnt parameters are necessary to be stored as chunks so that they are easy to recollect.

Ans 3. In my view, reward is the central necessary entity that determines whether a person would remain associated with some task or completely leave it. This might not be of physical nature. It might just be an abstract notion that gives phychological satisfaction or internal motivations (release of hormones etc.) that reinforce an action.
In case of a fire-fighter, he initially tries out the knowledge that he has gained during his training, from others' experience. He has some basic idea as to how to complete his task. Based on how the task is rewarded (positive or negative), he forms abstractions to carry out his work. The reward or the reinforcement in this case might include the advise or the response he gets from his peers, whether the fire was put out and his internal satisfaction in carrying out the task.

References:

  1. Kalakrishnan, Mrinal, et al. "Learning force control policies for compliant manipulation." Intelligent Robots and Systems (IROS), 2011 IEEE/RSJ International Conference on. IEEE, 2011.