New Method Scales Observation-Aware Planning in Uncertain Domains
A new method addressing the Optimal Observability Problem (OOP) enhances planning in uncertain domains by weighing task achievability against hardware and processing costs. Building on the Partially Observable Markov Decision Process (POMDP) model, the study investigates sub-symbolic techniques for decidable fragments, specifically the Sensor Selection Problem (SSP) and Positional Observability Problem (POP). This innovative approach decomposes POMDPs to pinpoint effective observation functions, yielding improvements of three orders of magnitude in instance size and five in runtime compared to original parameter synthesis methods. The research falls under Computer Science, focusing on Artificial Intelligence.
Key facts
- The Optimal Observability Problem (OOP) balances task achievability against hardware and processing costs.
- OOP is based on the Partially Observable Markov Decision Process (POMDP) model.
- The work studies sub-symbolic techniques for decidable fragments: Sensor Selection Problem (SSP) and Positional Observability Problem (POP).
- A new solving method decomposes POMDPs to identify sensible observation functions.
- Performance improved by 3 orders of magnitude for instance size and 5 orders of magnitude for runtime.
- The original approach was based on parameter synthesis.
- The problem involves deciding which sensing capabilities to deploy on an agent.
- The research is categorized under Computer Science > Artificial Intelligence.
Entities
—