Managing an Agent's Changing Intentions Using LTLfSynthesis

Autonomous agents' intentions (goals they are committed to) typically change as they operate. We develop a new model of intention change for such agents. We assume that the agent operates in a fully observable nondeterministic (fond) domain and uses Linear Temporal Logic over finite traces (ltl 𝑓) to represent intentions. We exploit ltl 𝑓 synthesis notions and techniques to generate strategies for the agent to satisfy its intentions and to revise them when the agent adopts new intentions or drops existing ones; this ensures that the agent's intentions always remain realizable. We propose automata-based methods to efficiently manage ltl 𝑓 intentions by exploiting auxiliary data structures built during synthesis. We implement a prototype and evaluate its effectiveness experimentally.