Location Aware Resource Management in Smart Homes Abhishek Roy, Soumya K. Das Bhaumik, Amiya Bhattacharya, Kalyan Basu, Diane J. Cook and Sajal K. Das CReWMaN Lab, Computer Science and Engineering Department The University of Texas at Arlington aroy, sdas, bhatt, basu, cook, das @cse.uta.edu Abstract— The rapid advances in a wide range of wireless access tech- nologies along with the efficient use of smart spaces have already set the stage for development of smart homes. Context-awareness is perhaps the most salient feature in these intelligent computing platforms. The “loca- tion” information of the users plays a vital role in defining this context. To extract the best performance and efficacy of such smart computing envi- ronments, one needs a scalable, technology-independent location service. In this paper we have developed a predictive framework for location-aware resource optimization in smart homes. The underlying compression mech- anism helps in efficient learning of an inhabitant’s movement (location) profiles in the symbolic domain. The concept of Asymptotic Equipartition Property (AEP) in information theory helps to predict the inhabitant’s fu- ture location as well as most likely path-segments with good accuracy. Suc- cessful prediction helps in pro-active resource management and on-demand operations of automated devices along the inhabitant’s future paths and lo- cations — thus providing the necessary comfort at a near-optimal cost. Sim- ulation results on a typical smart home floor plans corroborate this high prediction success and demonstrate sufficient reduction in daily energy- consumption, manual operations and time spent by the inhabitant which are considered as a fair measure of his/her comfort. I. I NTRODUCTION The essence of Weiser’s ubiquitous computing vision [20] lies in the creation of smart environments saturated with comput- ing and communication capabilities, yet gracefully integrated with human users. The two distinctive characteristics of ubiq- uitous computing are: (i) the noticeable migration of comput- ing from general-purpose computers to smaller customized mo- bile terminals, and (ii) the pro-active interaction and inherent sentience [11] of the computing devices with their surrounding network infrastructure. “Context-awareness” is perhaps the key characteristic of next-generation intelligent networks and asso- ciated applications. Location awareness is the most important “context” for the vast majority of ubiquitous computing scenar- ios, since the information needed by users depend strongly on their current or near future location. A quick look into different such applications like Advanced Traveler Information Systems (ATIS) [19], electronic tourist guides [1] and fleet management systems [21] reveal that the prime objective of all these proto- types is to improve the convenience of the visitors. This vision of ubiquitous computing has already given birth to a new research area: ‘intelligent location management in smart indoor environments.’ In such an environment the technology needs to be weaved into the fabric of our everyday life such that it becomes “technology that disappears” [20]. Over the past few years, there has been an upsurge of several innovative proto- types for indoor location-aware computing platforms. The Ac- tive Badge [8] is perhaps the first infra-red based location track- ing system developed for indoor offices. Active Bat [9] takes This work was supported by NSF grants under EIA-0086260, EIA-0115885 and IIS-0121297 and Texas Telecommunications Engineering Consortium (Tx- TEC). the help of ‘ultrasonic time-of-flight lateration technique’ to im- prove the granularity of location-sensing offered by the badge system. On the other hand, MIT’s Cricket Location Support System [18] delegates the responsibility of location reporting to the mobile object itself. RADAR [3], another RF-based indoor location support system uses signal strength and signal-to-noise ratio to compute 2-D positioning. The Motion Star [2] track- ing system uses electromagnetic sensing and virtual reality to compute the required location information. Microsoft’s Easy- living and Microsoft Home [13] projects use real-time 3D cam- eras to provide stereo-vision positioning capability in an indoor environment. In the Georgia Tech’s Aware Home [17], the em- bedded pressure sensors capture inhabitant’s footfalls, and the system uses this data for position tracking and pedestrian recog- nition. The Neural Network House project [16] of the University of Colorado focuses on the development of an adaptive control of home environments (ACHE) to anticipate the needs of the in- habitants. The Intelligent Home Project [14] at the University of Massachusetts explores the application of multi-agent systems technology to develop and maintain a smart indoor environment. MIT’s Intelligent House n [12] also focuses on developing ex- cellent products and services to satisfy the needs of the people living in the future-generation houses. A careful insight into the features of these location services reveals that the ability to predict the inhabitant’s future location often becomes the key to system’s associated “smartness”. In this paper we have taken an information-theoretic approach to develop a location-aware resource optimization scheme in the smart home environments. The complexity of the indoor location management is related to entropy, a well-known mea- sure for uncertainty of a probabilistic source. An analysis of inhabitant’s daily routine reveals some patterns in his/her daily- life. Although the life style changes over time, but such changes are not frequent and random. This observation helps us to as- sume the inhabitant’s mobility as a piece-wise stationary, er- godic, stochastic process. Exploiting this general, yet realistic assumption, the LeZi update [4], [5] provides an asymptotically optimal location information in the symbolic domain. Although, in a smart home, there exists a wide number of possible routes from one part of the room to the another, an inhabitant usually follows his/her most likely paths. A similar analogy, dealing with the asymptotic equipartition property (AEP) [22] in infor- mation theory, states that among all the long-range sequences consisting of random variables, there exists a fairly small typ- ical set [22] which contains most of the probability mass and controls the average behavior of all such sequences. Using this concept, one can capture the inhabitant’s typical path segments. We claim that reserving resources and activating the automated