-
Notifications
You must be signed in to change notification settings - Fork 72
BakSTP
The notion of time is central to temporal planning. EUROPA uses variables to explicitly represent timepoints for plan activities and states. Constraints among timepoints provide a natural way to express domain axioms. For example, in order to state that activity A must occur before activity B we can say that the end timepoint of A is <= the start timepoint of B. Dechter et al. (Dechter 1991) proposed that constraints among timepoints can be grouped together to form a Simple Temporal Network (STN). Such a network can be transformed into a Distance Graph (DG) where the outward arc from a node represents the maximum distance from the source node to the target node. The diagram below illustrates a simple STN with just 2 variables and a single constraint. It also shows the resulting DG.
![stn] (images/stn.png)
Dechter et al. (Dechter 1991) also showed that shortest path algorithms could be used to propagate values in the network and discover a negative cycle. A negative cycle is a path from a node to itself that has a path length less than 0. If such a cycle exists, the network is inconsistent. It was further shown that a single-source shortest path algorithm was sufficient to detect a negative cycle and provide sufficient propagation to yield a backtrack-free search. Thus we have an efficient and complete algorithm for propagating an STN. These results build on the already established notion of a CSP and are naturally incorporated into the general representation and propagation scheme used in EUROPA.