0018-9162/01/$10.00 © 2001 IEEE50 ComputerImplementing a Sentient Computing SystemAs computer users become increasinglymobile, and the diversity of devices withwhich they interact increases, so the over-head of configuring and personalizing thesesystems increases. A natural solution to thisproblem would be to create devices and applicationsthat appear to cooperate with users, reacting asthough they are aware of the context and manner inwhich they are being used, and reconfiguring them-selves appropriately.At AT&T Laboratories Cambridge, we have builta system that uses sensors to update a model of thereal world. We designed the model’s terms—objectpositions, descriptions and state, and so forth—to beimmediately familiar to users. Thus, the modeldescribes the world much as users themselves would.We can use this model to write programs that react tochanges in the environment according to the user’spreferences. We call this sentient computing becausethe applications appear to share the user’s perceptionof the environment.1Treating the current state of the environment as com-mon ground between computers and users providesnew ways of interacting with information systems.Metaphysical concerns aside, a sentient computing sys-tem doesn’t need to be intelligent or capable of form-ing new concepts about the world—it only needs to actas though its perceptions duplicate the user’s. Forexample, suppose a user picks up a wireless device thata sentient computing system manages. The systemseems to be aware that this event has occurred, and itautomatically configures the device to that user.In earlier work, we described a prototype of this sys-tem and stated our intention to deploy it on a largescale.2We have now installed an enhanced versionthroughout an office building. Over the past year,approximately 50 staff members have used the systemdaily with a set of trial applications.IMPLEMENTING A SENTIENT SYSTEMOur project implemented the sentient computingsystem’s model of the world as a set of softwareobjects that correspond to real-world objects. Objectsin the model contain up-to-date information aboutthe locations and state of the corresponding real-world objects. Ascertaining object positions with near-human levels of accuracy requires a specially designedsensor system.Location sensingThe location sensor, shown in Figure 1, determinesthe 3D positions of objects within our building in realtime. Personnel carry wireless devices known as Bats,which can also be attached to equipment. The sensorsystem measures the time it takes for the ultrasonicpulses that the Bats emit to reach receivers installed inknown, fixed positions. It uses these times of flight tocalculate the position of each Bat and hence the posi-tion of the object it tags by triangulation.To allow accurate time-of-flight measurements, awireless, cellular network synchronizes Bats with theceiling receivers. Base stations simultaneously addressa Bat over the wireless link and reset the receivers overa wired network. A wireless back channel supports theBat’s transmission of registration, telemetry, and con-trol-button information.Current embodimentWe have installed the Bat system throughout ourthree-floor, 10,000-sq.-ft. office building, and all 50Sentient computing systems, which can change their behavior based on a model of the environment they construct using sensor data, may hold the key to managing tomorrow’s device-rich mobile networks.MikeAddleseeRupertCurwenSteve HodgesJoe NewmanPete StegglesAndy WardAT&T LaboratoriesCambridgeAndy HopperAT&T LaboratoriesCambridgeand Universityof Cambridge COVER FEATUREstaff members use it continuously. The system uses750 receiver units and three radio cells, and tracks200 Bats.Figure 2 shows the current Bat device. Each Batmeasures 8.0 cm × 4.1 cm × 1.8 cm, has a unique 48-bit ID, and draws power from a single AA lithium cell.In addition to two input buttons, a Bat has a buzzerand two LEDs for output. Applications send com-mands over the wireless network to generate feedbackvia these devices.Built around a DSP microprocessor, the locationsystem receiver units use a matched-filter signal-detec-tion approach. Receivers are wired together in a daisy-chain network and are installed unobtrusively abovethe tiles in our building’s suspended ceilings.To simplify maintenance of the location system,telemetry data can be obtained from the Bats, indi-cating current battery health and firmware versionnumber. Further, system administrators can send com-mands over the wireless and wired networks to repro-gram Bats and receivers in the field.About 95 percent of 3D Bat position readings areaccurate to within 3 cm. Each base station can addressthree Bats simultaneously, 50 times each second, giv-ing a maximum location update rate across each radiocell of 150 updates per second. The signals from simul-taneously triggered Bats are encoded using a differ-ential-phase modulation scheme, allowing receiversto distinguish among them.Scheduling and power savingBecause we use Bats to tag many mobile objects inthe environment, we must fully exploit the limitednumber of location-update opportunities. We alsotake every opportunity to reduce the power con-sumption of Bats because frequently changing the bat-teries of several hundred devices would requiresignificant effort.Base stations use a quality-of-service measure toshare location-update opportunities between Bats.The base stations can preferentially allocate locationresources to objects that move often or are likely tobe moved. A scheduling process that runs on eachbase station determines not only when the base sta-tion will address a particular Bat but also when it willnext address it. The process passes this informationto the Bat across the wireless link, allowing the Bat toenter a low-power sleep state for the intervening time.The scheduling algorithm and Bats incorporate mech-anisms that allow rapid changes to the QoS alloca-tions.3Rapid changes to the schedule prove particularlyuseful when a user presses a Bat button. This actiongenerally indicates that the user wishes to performsome task with the Bat. In these cases, it helps toobtain a low-latency position update for the Bat. TheBat transmits a button-press message over the wire-less network, and the base station responds by sched-uling an immediate location-update opportunity forthat Bat.Bats register with a base station using their unique48-bit IDs and receive a temporary 10-bit ID