Some architectural considerations
motters
that these will all be identical units. With the increasingly low
cost of digital cameras that would be quite feasible, and from an
engineering point of view it would make sense with parts being easily
interchangeable.
However, there may be situations where the vision system of the robot
is not so homogeneous. For example, on something resembling a Mars
rover, or even some kind of domestic robot you might have one stereo
camera with a relatively long baseline and wide field of view used for
navigation but you might also have a smaller stereo camera mounted on
the end of an arm used for close up inspection, which might have a
short baseline and narrow field of view. So I've added the ability to
have multiple sensor models within the Sentience system, with one
model being assigned to each stereo camera.
At the moment each sensor model lookup table is calculated numerically
from a simulation using a high resolution grid (1mm per cell). Almost
certainly there will be a more exact formula describing the
probability density function, but my mathematical skills at present
are insufficient to be able to work this out in closed form. Doing it
using a simulation also leaves some flexibility, so that if I've made
any incorrect assumptions the simulation could be changed and the
sensor models re-calculated in an easy way.