Proactive Assistance in Ecologies of Physically Embedded Intelligent Systems: A Constraint-Based Approach

Proactive Assistance in Ecologies of Physically Embedded Intelligent Systems: A Constraint-Based Approach

Marcello Cirillo (Örebro University, Sweden), Federico Pecora (Örebro University, Sweden) and Alessandro Saffiotti (Örebro University, Sweden)
DOI: 10.4018/978-1-61692-857-5.ch025


The main goal of this Chapter is to introduce SAM, an integrated architecture for concurrent activity recognition, planning and execution. SAM provides a general framework to define how an intelligent environment can assess contextual information from sensory data. The architecture builds upon a temporal reasoning framework operating in closed-loop between physical sensing and actuation components in a smart environments. The capabilities of the system as well as possible examples of its use are discussed in the context of the PEIS-Home, a smart environment integrated with robotic components.
Chapter Preview

1 Introduction

Intelligent environments are quickly evolving to incorporate sophisticated sensing and actuation capabilities thanks to the progressive inclusion of devices and technologies from the field of autonomous robotics. Several approaches are currently under way that combine insights from the fields of intelligent environments and autonomous robotics, and define different types of smart robotic environments in which several pervasive robotic devices cooperate to perform possibly complex tasks. Instances of this paradigm include network robot systems (Sanfeliu, Hagita, & Saffiotti, 2008), intelligent spaces (Lee & Hashimoto, 2002), sensor-actuator networks (Dressler, 2006), ubiquitous robotics (Kim, Kim, & Lee, 2004), artificial ecosystems (Sgorbissa & Zaccaria, 2004), and PEIS-Ecology (Saffiotti, et al., 2008). In this chapter, we generically refer to a system of this type as a “smart robotic environment”.

In order to provide personalized services to the user, intelligent environments in general and smart robotic environments in particular must employ non-trivial and temporally contextualized knowledge regarding the user’s state (Bjorn, Guesgen, & Hubner, 2006). For instance, if a smart home could recognize that the human user is cooking, it could avoid sending a robot to clean the dining room until the subsequent dining activity is over.

The problem we tackle in this chapter is that of realizing a service-providing infrastructure for proactive human assistance in smart robotic environments. Two key capabilities that are often desirable in such intelligent environments are (1) the ability to recognize activities performed by the human user, and (2) the ability to plan and execute the behavior of pervasive devices according to the indications of activity recognition.

This chapter presents SAM1, a reasoning infrastructure that provides these two key features through the use of temporal constraint reasoning. SAM adopts a knowledge-driven approach to activity recognition, in which requirements expressed in the form of temporal constraints are used to specify how sensor readings correlate to particular states of a set of monitored entities in the environment. These entities can represent, for instance, various levels of abstraction of a human user's daily activities. Also, through the same constraint based formalism, SAM provides a means to specify how the states of monitored entities should lead to the synthesis and execution of plans for service providing actuators. SAM continuously refines its knowledge on the state of affairs in the environment through an on-line inference process. This process concurrently performs abductive inference and proactive plan synthesis, the former providing the capability to recognize context from current and past sensor readings, the latter providing proactive and contextualized service execution.

SAM is instantiated as part of the PEIS-Home, a testbed environment based on the concept of PEIS-Ecology (Saffiotti, et al., 2008). In a PEIS-Ecology, robotic, sensory and intelligent software components are combined to obtain a context-aware, service-providing environment for the future home. SAM leverages the capabilities of the environment to sense the information required to perform abductive reasoning on user activities as well as its actuation capabilities to enact contextual services.

In the following, we first present the main concepts behind the PEIS-Ecology and its physical instantiation, the PEIS-Home. We then detail SAM, focusing first on how requirements for the abductive reasoning and proactive service planning can be expressed as temporal constraints, and then on how activity recognition and service enactment is performed through SAM's continuous inference process. We also present several case studies in the PEIS-Home, aimed at providing a proof-of-concept validation of the architecture. All case studies are obtained as a result of a human test subject performing daily activities within the PEIS-Home. Finally, we provide a summary of related work, providing examples of other approaches used to obtain similar functionalities as well as architectures similar to SAM employed in different application contexts.

Key Terms in this Chapter

Activity Recognition: The ability of the intelligent system to infer temporally contextualized knowledge regarding the state of the user on the basis of a set of heterogeneous sensor readings.

Decision Network: A network of constraints between decisions, where decisions represent either sensor readings or human activities.

Temporal Propagation: Propagation of temporal constraints (Temporal Propagation) in a Decision Network is used to assess the temporal consistency of hypotheses inferred on human activities.

Robot: In the PEIS-Ecology approach, the concept of robot is abstracted by the uniform notion of a PEIS, a physical device which includes a number of functional components.

Temporal Constraint: A temporal relation between variables that represent events in time. In this work, we consider relations in Allen’s Interval Algebra.

Smart Robotic Environment: An environment where robots, pervasive sensors and actuators cooperate to provide services for human users.

Abductive Inference: A method of logical inference that allows to infer preconditions from consequences.

PEIS-Ecology: An ecological approach to creating Smart Robotic Environments, which combines concepts from Artificial Intelligence, Robotics and Ubiquitous Computing. The power of a PEIS Ecology does not come from the individual power of its constituent PEIS, but it emerges from their ability to interact and cooperate.

PEIS: A “Physically Embedded Intelligent System” is a physical device which includes a number of functional components. Each PEIS can use functionalities from other PEIS in the ecology in order to compensate or to complement its own.

Complete Chapter List

Search this Book: