I would like to suggest that, whatever the outcome of the perception
proposal, we move the perception code into a seperate module in the
simulator. At present the kernel does all the computation for deciding
who can see (and hear) what. This is not such a good design as it
requires changing the kernel every time we want to modify the way
perception works.
I propose that we add a new layer to the kernel achitecture: perception
modelling. This will work in exactly the same way as the simulator
layer. The kernel will send the updates it receives from the simulators
to the perception module which replies with a set of SENSE messages
reflecting the changes that each agent can sense. In a similar way, a
communication module will turn AK_SAY and AK_TELL messages into KA_HEAR,
KA_HEAR_SAY and KA_HEAR_TELL messages.
With this architecture we can easily change the models of perception and
communication in the same way we currently swap fire simulators.
Cameron.
This archive was generated by hypermail 2.1.3 : Mon 15 Nov 2004 - 21:23:53 GMT