The most interesting part of day 3 for me was the session on location-aware applications. The first paper was The potential for location-aware power management from Robert Harle and Andy Hopper of the University of Cambridge. They analyze a set of highly-accurate location data from a commercial office to see what energy savings could be enabled by knowing where the user is located in the building. They found the potential for savings for devices that can be switched on and off quickly (like lights or waking a computer from sleep), but the problem is much harder for slower systems like HVAC. For the slow systems, only outside office hours and lunchtime could be predicted with the accuracy required to have rooms ready when the user returned.
The other big paper for me was Understanding mobility based on GPS data, which covered using GPS data to determine what transportation method was being used (sound familiar? 🙂 ) They found that just using velocity data from GPS doesn’t work well (< 50% accuracy). They follow the standard machine learning process: segmentation of the data, extracting features that are representative of those segments, and feeding that into a generative model. They had 70 people using their system for a year, which is a huge amount of GPS data. This is definitely a paper I need to fully understand.
I had to leave to catch my flight before the closing panel session finished, but what I heard was interesting. There was spirited debate on where ubiquitous computing was going, where the conference was going, complaints about the many related conferences (Pervasive, MobileCHI). Too bad I had to leave early.
Ubicomp 2009 is in Orlando FL. Not sure at this point if I’ll be going or not.