Fast and Complete Symbolic Plan Recognition

Recent applications of plan recognition face several open challenges: (i) matching observations to the plan library is costly, especially with complex multi-featured observations; (ii) computing recognition hypotheses is expensive. We present techniques for addressing these challenges. First, we show a novel application of machine-learning decision-tree to efficiently map multi-featured observations to matching plan steps. Second, we provide efficient lazy-commitment recognition algorithms that avoid enumerating hypotheses with every observation, instead only carrying out bookkeeping incrementally. The algorithms answer queries as to the current state of the agent, as well as its history of selected states. We provide empirical results demonstrating their efficiency and capabilities.