Search

This article provides an overview of techniques to deal with constraints by expressing planning within general constraint-solving frameworks. Using constraints and related techniques as an underlying framework for problem-solving tasks has proven successful for many domains. The basic modeling units of constraint-based frameworks are constraints and variables. Such frameworks can easily express complex problem features. To remain efficient most constraint-based search frameworks use only a restricted scenario.

This article proposes an approach that addresses the challenges in recognizing activities of daily living (ADL) and shows promise in automating some types of ADL monitoring. The system has three components: specialized sensors to detect object interactions, a probabilistic engine that infers activities given observations from sensors and a model creator that easily creates probabilistic models of activities. Radio frequency identification tags' durability and their ability to be placed out of sight make them particularly well suited for ADL monitoring of elders with early-stage dementia, who tend to disassemble foreign objects. PROACT represents activities as linear sequences of activity stages. The probabilistic engine converts these activity models into dynamic Bayesian networks. Activity referencing requires a model that defines an activity: how many stages it has and how sensor data determines the transitions between stages. Of the more than 20 ADL that caregivers can monitor, 14 were chosen for evaluation. The object-touch sequence for each subject, along with models for the 14 activities, were given to the inference engine. PROACT correctly inferred that an activity occurred 88 percent of the time. For eight ADL, there were no false positives. Activities with starting points posed a more subtle problem. INSETS: Activities of Daily Living;Related Work in ADL Inferencing.