Figure 10.8 shows an extension of Dijkstra's algorithm for solving the problem of Formulation 10.1 under nondeterministic uncertainty. It can also be considered as a variant of the algorithm in Figure 10.6 because it grows by using backprojections. The algorithm in Figure 10.8 represents a backward-search version of Dijkstra's algorithm; therefore, it maintains the worst-case cost-to-go, , which sometimes becomes the optimal, worst-case cost-to-go, . Initially, for states in the goal, and for all others.
Step 1 performs the initialization. Step 2 selects the state in that has the smallest value. As in Dijkstra's algorithm for deterministic problems, it is known that the cost-to-go for this state is the smallest possible. It is therefore declared in Step 3 that , and is extended to include .
Step 4 updates the costs for some states and expands the active set, . Which costs could be immediately affected by the insertion of into ? These are states for which there exists some that produces a one-stage forward projection, , such that: 1) and 2) . This is depicted in Figure 10.9. Let the set of states that satisfy these constraints be called the frontier set, denoted by . For each , let denote the set of all actions for which the forward projection satisfies the two previous conditions.
The frontier set can be interpreted in terms of backprojections. The weak backprojection yields all states that can possibly reach in one step. However, the cost-to-go is only finite for states in (here already includes ). The states in should certainly be excluded because their optimal costs are already known. These considerations reduce the set of candidate frontier states to . This set is still too large because the same action, , must produce a one-stage forward projection that includes and is a subset of .
The worst-case cost-to-go is computed for all as
Steven M LaValle 2020-08-14