Logic-Based Approaches to Intention Recognition

1 Introduction

Intention recognition, also called goal recognition, is the task of recognizing the intentions of an agent by analyzing some or all of their actions and/or analyzing the changes in the state (environment) resulting from their actions. Plan recognition is closely related to intention recognition, and extends it to recognizing the plan (i.e. the sequence of actions, including future actions) the observed agent is following in order to achieve his intention. Throughout this paper we will use goal and intention interchangeably.

Work on intention recognition has been going on for about 30 years. Examples of early work are attributed to Schmidt et al. [1978], Wilensky [1983], and Kautz and Allen [1986]. Much of the early work has been in the context of language and story understanding and automatic response generation, for example in Unix help facilities. However, new applications, such as assisted living and ambient intelligence, increasingly sophisticated computer games, intrusion and terrorism detection, and the military have brought new and exciting challenges to the field. For example assisted living applications require recognizing the intentions of residents in domestic environments in order to anticipate and assist with their needs. Applications in computer systems intrusion or terrorism detection require recognizing the intentions of the would-be-attackers in order to prevent them. Military applications need to recognize the intentions of the enemy maneuvers in order to plan counter-measures and react appropriately. Programs that make moral decisions (e.g. Pereira and Saptawijaya 2009) need to reason about intentions, in particular to decide whether untoward consequences of some actions were intended by the agent that performed them or were merely unintended side-effects.

Logic has been a powerful tool in intention recognition. Since the early days, for example in the work of Charniak and McDermott [1985], abduction has been used as the underlying reasoning mechanism in providing hypotheses about intentions. Also, conceptually, intention recognition is directly related to planning, and logic has been the basis of many causal theories describing the relationship between actions and effects.

Intention recognition is a rich and challenging field. Often multiple competing hypotheses are possible regarding the intentions of an observed agent. The choice between these hypotheses is one challenge, but there are many others. One, for example, is that circumstances, including the adversarial nature of the observed agent, may afford only partial observability of the actions. Geib and Goldman [2001] make a contribution in this respect, as do Sindlar et al [2008], described in Section 4 in this article. Furthermore, would-be-intruders and would-be attackers may even deliberatively execute misleading actions.

Another challenge is the case where the acting agent may have multiple intentions and may interleave the execution of his plans of actions for achieving them, or the case where the actor is concurrently trying alternative plans for achieving the same intention. Intention recognition becomes more difficult when we attempt to interpret the actions of cognitively impaired individuals who may be executing actions in error and confusion, for example in the case of Alzheimer patients (Roy et. al. 2007). Similar complications arise and are magnified when attempting to analyze the actions and intentions of multiple (co-operating) agents (e.g. Sukthankar and Sycara 2008).

In this article we will explore the field of intention recognition, and in particular we will focus on single agent cases, as opposed to multi-agents, and on logic-based approaches. We will explore the logical basis of intention recognition, and provide an analysis of eight case studies. The case studies are chosen from the literature for their variety of methodologies and applications. We will not consider many probabilistic approaches to intention recognition, except in two of the case studies, Pereira and Anh [2009b] and Demolombe and Frenandez [2006], both of which combine logic with probabilities.

In Section 2 we look at the background and issues involved in intention recognition. In Section 3 we look at the possible relationships between logic-based causal theories and knowledge representation and reasoning for intention recognition. In Section 4 we describe and analyze the eight case studies. Finally, in Section 5, we conclude with a further discussion of the challenges.