The task of keyhole (unobtrusive) plan recognition is central to adaptive game AI. "Tech trees" or "build trees" are the core of real-time strategy (RTS) game strategic (long term) planning. This paper presents a generic and simple Bayesian model for RTS build tree prediction from noisy observations, which parameters are learned from replays (game logs). This unsupervised machine learning approach involves minimal work for the game developers as it leverage players' data (com- mon in RTS). We applied it to StarCraft1 and showed that it yields high quality and robust predictions, that can feed an adaptive AI.
It is important for agents to model other agents' unobserved plans and goals, based on their observable actions, a process known as plan recognition. Plan recognition often takes the form of matching observations of an agent's actions to a planlibrary, a model of possible plans selected by the agent. In this paper, we present efficient algorithms that handle a number of key capabilities implied by plan recognition applications, in the context of hybrid symbolic-probabilistic recognizer. The central idea behind the hybrid approach is to combine the symbolic approach with probabilistic inference: the symbolic recognizer efficiently filters inconsistent hypotheses, passing only the consistent hypotheses to a probabilistic inference engine. There are few investigations that utilize an hybrid symbolic-probabilistic approach. The advantage of this kind of inference is potentially enormous. First, it can be highly efficient. Second, it can efficiently deal with richer class of plan recognition challenges, such as recognition based on duration of behaviors, recognition despite intermittently lost observations, and recognition of interleaved plans.
One of the most important problems for an intelligent tutoring system is deciding how to respond when a student asks for help. Responding cooperatively requires an understanding of both what solution path the student is pursuing, and the student's current level of domain knowledge. Andes, an intelligent tutoring system for Newtonian physics, refers to a probabilistic student model to make decisions about responding to help requests. Andes' student model uses a Bayesian network that computes a probabilistic assessment of three kinds of information: (1) the student's general knowledge about physics, (2) the student's specific knowledge about the current problem, and (3) the abstract plans that the student may be pursuing to solve the problem. Using this model, Andes provides feedback and hints tailored to the student's knowledge and goals.
This paper presents an anytime algorithm for incremental intention recognition in a changing world. The algorithm is performed by dynamically constructing the intention recognition model on top of a prior domain knowledge base. The model is occasionally reconfigured by situating itself in the changing world and removing newly found out irrelevant intentions. We also discuss some approaches to knowledge base representation for supporting situation-dependent model construction. Reconfigurable Bayesian Networks are employed to produce the intention recognition model.
Plan recognition does not work the same way in stories and in "real life" (people tend to jump to conclusions more in stories). We present a theory of this, for the particular case of how objects in stories (or in life) influence plan recognition decisions. We provide a Bayesian network formalization of a simple first-order theory of plans, and show how a particular network parameter seems to govern the difference between "life-like" and "story-like" response. We then show why this parameter would be influenced (in the desired way) by a model of speaker (or author) topic selection which assumes that facts in stories are typically "relevant".