17.01.2013 Views

Chapter 2. Prehension

Chapter 2. Prehension

Chapter 2. Prehension

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

188 THE PHASES OF PREHENSION<br />

A +6 1<br />

A-6 2<br />

- . Initial Position Ballpark<br />

OfA<br />

Movefast Schema:<br />

Move to some<br />

location<br />

near the target A<br />

Moveslow Schema:<br />

Move towards and<br />

through A, and<br />

expect another<br />

schema to be<br />

triggered<br />

Figure 5.29 Arbib, Iberall, Lyons (1985) notion of ‘ballpark’. A)<br />

Placing a mug on a table involves getting the mug to a location<br />

above the target location, and then moving the mug through the<br />

target location (i.e., through the table), resulting in sensory<br />

contact triggering other schemas. B) Generalization of the ballpark<br />

model showing how two Move Schemas work together.<br />

For the fist phase of a movement, adding to the goal position A ,<br />

would prevent accidental contact with the object; for the second phase<br />

of a movement, subtracting & would ensure contact, as the goal of the<br />

movement would actually be inside the object. In this way, uncer-<br />

tainty can be dealt with in perceptual data.<br />

These reaching schemas can be modelled using neural networks in<br />

terms of ballpark modelling. Recalling the Coordinated Control<br />

Program (Figure 5.1), the hand is moved by the Movefast Schema<br />

from a given starting location and posture to be within the ballpark of a<br />

desired location. The Reshape Schema shapes the virtual fingers into<br />

a posture larger than the object at peak aperture. Then, the Moveslow<br />

Arm Schema moves the wrist beyond the desired location. The<br />

Enclose Schema brings the virtual fingers together, bringing the<br />

fingers ‘into’ the object (of course, the object surfaces stop the fingers<br />

at the appropriate time). The Approach Vector Selector model,<br />

described in <strong>Chapter</strong> 4, provides one method for computing a goal<br />

hand location (Iberall, 1987a). This ‘winner take all’ non-adaptive<br />

framework computed a good location for contacting a 4 cm dowel with<br />

two virtual fingers in pad opposition. A good location for the hand is<br />

one where the VFs could reach the ends of the dowel, and where the<br />

pads are aligned with each other as well as with the dowel. Following

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!