Coordinating Heterogeneous Agents to Synthesize Proactive Monitoring Amedeo Cesta, Gabriella Cortellessa, Federico Pecora, and Riccardo Rasconi ISTC-CNR, National Research Council of Italy Institute for Cognitive Science and Technology Via S.Martino della Battaglia 44, I-00185 Rome, Italy <name.surname>@istc.cnr.it Abstract This paper describes the results of the ROBOCARE, a project aimed at creating an integrated environment endowed with heterogeneous software and robotic agents for assisting an elderly person at home. Specifically, a proactive environment for continuous daily activity monitoring has been created in which an autonomous robot acts as the main interactor with the person. This paper describes how the synergy of different component technologies guarantees an overall intelligent be- havior capable of personalized and contextualized interaction with the assisted person. Environment for continuous monitoring In the ROBOCARE project we have been dealing with the problem of monitoring an older person during her daily ac- tivity at home. The use of intelligent technology for support- ing older people at home has been addressed in various re- search projects in the last years, e.g., PEAT (Levinson 1997), PEARL (Pineau et al. 2003; Pollack 2005), I.L.S.A. (Haigh, Kiff, & Ho 2006). In ROBOCARE we have followed the specific direction of creating a home environment dedicated to monitoring a person rather than concentrating all the functionalities on a single robot. The result is a prototypical intelligent environ- ment that integrates robotic and software components to ob- tain a continuous behavior that we call here Proactive Moni- toring. The goal underlying the intelligent environment con- cerns the ability to (a) maintain continuity of behavior, such as ensuring continuous monitoring of what happens in the environment (the state of the assisted elder and of his/her do- mestic context), (b) create a context at the knowledge level around the actions that the assisted elder performs (routinely, exceptionally, etc.), and (c) provide contextualized interac- tion services with the assisted elder aimed at proactive as- sistance. In particular this paper describes how we have achieved continuous proactive monitoring in a domestic environment, that is, a specific implementation of the sense-plan-act cycle, by integrating separate intelligent components as a multi- agent system. A coordination algorithm guarantees a consis- tent collective behavior of the entire environment by contin- uously solving a Distributed Constraint Optimization Prob- lem (DCOP). The coordination problem provides the “se- mantic glue” of the system, i.e., its resolution orients the en- vironment toward guaranteeing safety of the observed per- son. In this paper we describe how we have obtained this comprehensive behavior by integrating: (1) the use of a distributed constraint optimization algorithm for coordina- tion of the multiple agents involved in activity monitoring; (2) the use of a constraint based scheduling system, and in particular of the continuous schedule monitoring function- ality employed to provide appropriate alerts and warnings at the occurrence of constraint violations; (3) the genera- tion of relevant explanations from these constraint violations to be presented to the user and other observers of the pro- cess in the form of verbal interaction instances. This paper specifically focuses on the aspects related to the system’s context-awareness and interaction capabilities. In particular, we describe how the constraint-based scheduling technology is used to maintain a knowledge repository aimed at sup- porting on-demand specific interactions as well as enabling autonomous system initiative. Separate intelligent capabilities The main “actor” in our smart home environment is a robotic agent with verbal interaction capabilities. The robot acts as a “mediator” through which the assisted person receives ad- vice/warnings and can query the environment. As shown in Figure 1, the robot is composed of two distinct mod- ules. The mobile robotic platform provides advanced mo- bility functionalities (referred to as “robot motion skills” in the figure 1 ). A second module creates an additional level of competence for the robot, referred to as “interactive skills” in Figure 1(a). Indeed this capability groups and provides access to the functionalities of the overall intelligent sys- tem. Figure 1(b) shows how the interaction skills use (a) a simplified Interaction Manager, (b) a front-end for the inter- action module consisting in a Talking Head and a Speech Recognition subsystem taken as external off-the-shelf mod- ules, (c) a key component called Intelligent Activity Monitor whose role is very relevant for the situated interaction capa- bility of the system. 1 The robotic platform, developed by colleagues from Univer- sity of Rome “La Sapienza”, consists of a Pioneer 2 integrated with a Sick laser scanner for localization. Additional work has been required to both integrate advanced SLAM algorithms and obtain robust navigation abilities which are suited for the domestic environment. Details are outside the scope of the paper.