Fault Tolerant Planning: Toward Probabilistic Uncertainty Models in Symbolic Non-Deterministic Planning

Symbolic non-deterministic planning represents action effects as sets of possible next states. In this paper, we move toward a more probabilistic uncertainty model by distinguishing between likely primary effects and unlikely secondary effects of actions. We consider the practically important case where secondary effects are failures, and introduce n-fault tolerant plans that are robust for up to n faults occurring during plan execution. Fault tolerant plans are more restrictive than weak plans, but more relaxed than strong cyclic and strong plans. We show that optimal n-fault tolerant plans can be generated by the usual strong algorithm. However, due to non-local error states, it is often beneficial to decouple the planning for primary and secondary effects. We employ this approach for two specialized algorithms 1-FTP (blind) and 1-GFTP (guided) and demonstrate their advantages experimentally in significant real-world domains.