Context-Aware Smart Spaces:

Our Vision:

Ubiquitous computing offers the vision of a physical environment that is fundamentally more responsive to people.  Users will be able to benefit from the “smartness” of rooms and public spaces to seamlessly interact with a wide variety of  internetworked wireless and wired devices, e.g. printers, monitors, speakers, wearable computers, appliances, kiosks, toys, sensors, other wireless personal digital assistants and other video-enabled mobile phones.  Emerging technologies such as wireless pico-cell Bluetooth networks, wireless Ethernet, third generation high speed cellular systems, low power high speed microprocessors, and matchbox-sized gigabyte disks are already hastening the arrival of a world in which smart spaces are truly pervasive.   

Research Papers:

Our Research: 

One of our initial goals was to address what we term the active device resolution problem in pervasive smart spaces.  A common mode of interaction with a device-filled environment is via remote control.   Given a user with a wireless PDA or cell phone who wishes to remotely control one of N devices or services within a smart room, which of the N devices is the intended or “active” device whose user interface (UI) should be automatically selected and prefetched before the user manipulates the remote control PDA?   This problem is illustrated in the figure below.

Automated selection of the UI eases the user’s interaction with a pervasive environment, in comparison to manual selection, which is cumbersome, and directional remote control, e.g. pointing and clicking, which is insufficient to resolve the active device ambiguity when a group of devices is closely spaced, as in a home entertainment system.  Our goal is to anticipate the next user interface desired by the user based on each user’s history of remote control accesses.  

We are currently collecting traces of user remote control behavior while utilizing a wireless PDA.  We are subjecting these traces to a variety of various A.I./machine learning prediction algorithms, including first and second order Markov prediction as well as naïve Bayesian prediction.  We have been able to obtain accuracies of prediction thus far of 75-90%, depending upon the algorithm and the degree of training.