Summary: Reactive Systems
notes by: Eyal Amir
March 4, 1997
Reactive Systems deals with systems that reason in a timedependent fa
sion. The various issues include architectures, algorithms for timedependent
planning, and adjustments of techniques in AI to robots engaged in realtime
1 Boddy & Dean Time dependent planning
Note: This paper does not really belong in the Reactive Systems part. It is
somewhat between planning and robot planning, and as such, closer to McDer
mott's paper about robot planning.
ExpectationDriven Iterative Refinement is a method introduced by Boddy
and Dean in 89', that uses some expectation over the improvement of extra
deliberation time t, to decide whether that extra time will be used for extra
planning. It relies on an anytime algorithm, and uses past experience to give
expectations about its improvement over time.
An example of an anytime algorithm is and algorithm for path planning. The
simplest plan is to do DFS till you get to the node you were looking for, say node
n. Then after a while a path is found to n, and we simplified our plan. Edge
exchange is another example of an anytime algorithm for tour improvement.