Let's do statistical models for non-deterministic online navigation (AIMA ch. 4)

4.11c asks you to develop an error-correction strategy for online search agents which can recover from multiple, sequential failures; this brings up some interesting issues: it turns out that it might be necessary to collect statistics on transitions to determine what the most likely result of an action might be.

In order to assess one's next move, furthermore, it may be necessary to calculate the expected utility as a sum of cost-probabilities.

Let's develop a planning agent that keeps track of state-transition probabilities, even during error correction; and that can take advantage of accidental loops during error-recovery. This may also involve a novel data-structure that behaves like a stack but with constant-time random access.