Location/time based reminder systems are technically more complex, requiring an additional channel of information: the user's location. How this information is acquired and how reliable it is differs from implementation to implementation, with consequences for the reliability and scope of the reminders which can be delivered. Examples of the way location context may be acquired include GPS receivers (outdoors only)[6], tagged environments and portable sensors[9], wearable tags and "smart" environments[11,5], computer vision[1,10], and combinations of other types of sensors[4]. Each of these techniques have specific strengths and weaknesses, the most important being the degree of infrastructure dependence.
Being able to condition reminders on location is a powerful addition to a proactive reminder system, enabling location-only conditions (e.g. "next time I'm at the grocery store, remind me to buy milk") as well as sophisticated location-time conditions, such as varying the lead-time of a reminder to go to a meeting depending on current location, time, and probable time of transit.
Location context is broadly useful because it is an important clue to user activity[8]. For example, very different activities take place at home, at the supermarket, and in the office; there is little point in reminding the user to go to a meeting or to go grocery shopping if the user is already engaged in these activities as indicated by the user's current location. However, location context by itself can not differentiate between multiple activities which take place in the same space.
For instance, the comMotion location-based proactive reminder and message delivery system developed by Natalia Marmasse and Chris Schmandt uses GPS to determine location [6]. Differential GPS location measurement has a high degree of accuracy outdoors but does not work indoors. Hence, the comMotion system can identify location at the building level but does not differentiate between being in one part of a building vs. another. This means reminders can be effectively targeted at activities which are segregated by location at the building level (grocery shopping vs. working out, for instance), but not at the room level (eating in the office cafeteria vs. meeting in the conference room).
Likewise, using a Locust-style active IR tag system[9] it is possible to do comparatively fine-grained location determination indoors, but no amount of location accuracy can differentiate between different events which occur in the same location. For instance, determining that one of the authors of this paper is sitting at his office workstation can not, by itself, differentiate between him writing his thesis or playing networked Quake II capture-the-flag. Knowing whether a user is working, resting, or in conversation with some other person has obvious implications for reminder delivery, yet all of these activities may take place at unpredictable times in the same cluttered office.