Dynamic Controllability with Single and Multiple Indirect Observations

Paul Morris (1) , Arthur Bit-Monnot (2)

1 ARC - NASA Ames Research Center
2 UNISS - Università degli Studi di Sassari = University of Sassari [Sassari]

A recent paper introduced a transformation-based approach for determining dynamic controllability of Simple Temporal Networks with Uncertainty (STNUs) extended to have variably-delayed observations of uncontrolled timepoints. Although the approach correctly determines dynamic control-lability, it does not always provide the most flexible possible dynamic strategy. We show how to refine the approach in a way that improves the flexibility, and further extend it to a class of Partially Observable STNUs where the hidden timepoints can be indirectly observed via a chain of contingent links. We show how to construct a labeled distance graph for these problems, leading to a complete solution. This approach handles "single-headed" chained contingent links. For "multi-headed" problems, we prove a theorem characterizing their dynamic controllability in isolation. This provides a check on more general networks (and more general methods). We also consider potential extensions of the single-headed approach to multi-headed problems and point out some difficulties that arise.

Domaines: Intelligence artificielle [cs.AI]

Paul Morris, Arthur Bit-Monnot. Dynamic Controllability with Single and Multiple Indirect Observations. ICAPS Workshop on Integrating Planning, Acting, and Execution (IntEx), Jul 2019, Berkeley, United States. ⟨hal-02935700⟩