When acting, agents may deviate from the optimal plan, either because they are not perfect optimizers or because they interleave multiple unrelated tasks. In this paper, we detect such deviations by analyzing a set of observations and a monitored goal to determine if an observed agent's actions contribute towards achieving the goal. We address this problem without pre-defined static plan libraries, and instead use a planning domain definition to represent the problem and the expected agent behavior. At the core of our approach, we exploit domain-independent heuristics for estimating the goal distance, incorporating the concept of landmarks (actions which all plans must undertake if they are to achieve the goal). We evaluate the resulting approach empirically using several known planning domains, and demonstrate that our approach effectively detects such deviations.

Goal Recognition Design (GRD) is the problem of designing a domain in a way that will allow easy identification of agents' goals. This work extends the original GRD problem to the Plan Recognition Design (PRD) problem which is the task of designing a domain using plan libraries in order to facilitate fast identification of an agent's plan. While GRD can help to explain faster which goal the agent is trying to achieve, PRD can help in faster understanding of how the agent is going to achieve its goal. we define a new measure that quantifies the worst-case distinctiveness of a given planning domain, propose a method to reduce it in a given domain and show the reduction of this new measure in three domains from the literature.

Computer-based human activity recognition of daily living has recently attracted much interest due to its applicability to ambient assisted living. Such applications require the automatic recognition of high-level activities composed of multiple actions performed by human beings in an environment. In this work, we address the problem of activity recognition in an indoor environment, focusing on a kitchen scenario. Unlike existing approaches that identify single actions from video sequences, we also identify the goal towards which the subject of the video is pursuing. Our hybrid approach combines a deep learning architecture to analyze raw video data and identify individual actions which are then processed by a goal recognition algorithm that uses a plan library describing possible overarching activities to identify the ultimate goal of the subject in the video. Experiments show that our approach achieves the state-of-the-art for identifying cooking activities in a kitchen scenario.

Prior work on viewing plan recognition as parsing of grammars has assumed completely observable actions. This paper provides an algorithm to rewrite plan grammars to allow for recognizing partially observable actions.  For the ELEXIR (Geib 2009) system, the impact of this rewriting on plan recognition runtime is shown to be limited to those plans that actually use the partially observable actions.

It may (2010) to estimate the probability of each possible goal be that one agent needs to monitor the activities of another based on the difference between the cost of the best plan agent, attempt to assist the other agent, or simply avoid getting for the goal given the observed actions, Cost(G O), and the in the way while performing its own duties. For all of cost of the best plan for the goal without the observed actions, these cases the agent needs to be able to realize what the Cost(G O). The big difference here is that the observations other agent is doing. In the absence of full and timely communication only indirectly give us probabilities for actions in of plans and goals, goal and plan recognition becomes the plan graph. We therefore first construct a Bayesian Network essential. Many goal recognition techniques allow the (BN) to estimate these action probabilities, and then sequence of observations to be incomplete, but few consider use this probability information in the plan graph to compute the possibility of noisy observations. In practice, this is not expected cost for each goal, given the observations.

We present a technique to further narrow the gap between recipe-based and domain theory-based plan recognition through decompositional planning, a planning model that combines hierarchical reasoning as used in hierarchical task networks, and least-commitment refinement reasoning as used in partial-order causal link planning. We represent recipes through decompositional planning operators and use them to compile observed agent actions into an incomplete decompositional plan that represents them; this plan can then be input to a decompositional planner to identify the recognized plan-space plan. Our model thus synthesizes the heretofore disparate recipe-based and domain theory-based plan recognition variants into a unified knowledge representation and reasoning model.

This course is a graduate level introduction to automatic discourse processing. The emphasis will be on methods and models that have applicability to natural language and speech processing. The class will cover the following topics: discourse structure, models of coherence and cohesion, plan recognition algorithms, and text segmentation. We will study symbolic as well as machine learning methods for discourse analysis. We will also discuss the use of these methods in a variety of applications ranging from dialogue systems to automatic essay writing.

This paper presents a new model of cooperative behavior based on the interaction of plan recognition and automated planning.  Based on observations of the actions of an "initiator" agent, a  "supporter" agent uses plan recognition to hypothesize the plans  and goals of the initiator.  The supporter agent then proposes and  plans for a set of subgoals it will achieve to help the initiator.  The approach is demonstrated in an open-source, virtual robot  platform.

This paper presents an ontology-based application for mobile devices which is responsible for supporting groups of people with the management of their shared tasks. The ontology stores the domain knowledge about collaborative tasks, which is used to support task recognition and relocation. Such knowledge is used by a multi-agent system that consists of a group of agents representing each person in the group. The agents use plan recognition techniques to monitor the execution of tasks according to the schedules and negotiate task allocation when needed. Our techniques have been applied in a healthcare scenario which consists of a family group that takes care of an elderly person. This paper presents an ontology-based application for mobile devices which is responsible for supporting groups of people with the management of their shared tasks. % in a healthcare scenario.The ontology stores the domain knowledge about collaborative tasks, which is used to support task recognition and relocation.Such knowledge is used by a multi-agent system that consists of a group of agents representing each person in the group.The agents use plan recognition techniques to monitor the execution of tasks according to the schedules and negotiate task allocation when needed.Our techniques have been applied in a healthcare scenario which consists of a family group that takes care of an elderly person.

In many real-world domains, the presence of machines is becoming more ubiquitous to the point that they are usually more than simple automation tools. As part of the environment amongst human users, it is necessary for these computers and robots to be able to interact with them reasonably by either working independently around them or participating in a task, especially one with which a person needs help. This interactive procedure requires several steps: recognizing the user and environment from sensor data, interpreting the user’s activity and motives, determining a responsive behavior, performing the behavior, and then recognizing everything again to confirm the behavior choice and replan if necessary. At the moment, the research areas addressing these steps, activity recognition, plan recognition, intent recognition, and planning, have all been primarily studied independently. However, pipelining each independent process can be risky in real-time situations where there may be enough time to only run a few steps. This leads to a critical question: how do we perform everything under time constraints? In this thesis summary, I propose a framework that integrates these processes by taking advantage of features shared between them.