*Important* --Readings for next week (and next topic)

[Subbarao Kambhampati <rao@asu.edu>]

For the "handling partial observability" topic to start next week, here is the

basic outline:

We will first consider "non-deterministic" partial observability (that is you, you know you 

are in one of K states, but  don't have a

distribution on the K states).   

 Within this context we will look at conformant (don't sense) and conditional (sense once in a while) planning problems. 

The actions themselves will be either deterministic or non-deterministic. 

***Russell & Norvig Section 4.4 will be the atomic model

    (You can also, alternatively, look at this paper: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.42.9524

***Section 2 in http://rakaposhi.eas.asu.edu/dan-jair-pond.pdf  will be the factored model

***The rest of that paper (especially Section 3 and 4) discuss ways of generalizing planning graph heuristics for this scenario.

=======

We will then look at stochastic partial observability (i.e., we have a distribution over the states). The atomic model in this case will be

POMDP 

***Russell & Norvig Section 17.4 will be the atomic model

(and if we survive this, we will consider factored models for POMDPs)

Rao