CABot2
- The EPSRC plan had three agents.
- I planned a little less work on CABot2, using some stuff from
CABot1.
- We also used CABot1 as a basis for some systems
that didn't make it into CABot2.
- None the less, CABot2 included an improved vision system. This
behaved sensibly when more than one object was in the visual
field.
- It also included an improved parser. I'll discuss an
improved
version of the CABot2 parser shortly.
- It also incorporated action-goal rules
to learn "center the pyramid".
- It is a lot faster (in simulation and real time), and only fails
on "go to the pyramid" (~10%). Here it fails when it overshoots, so
that could be judged a success.
