Augmented reality attempts to provide informative or entertaining overlays on the physical world. However, it is easy to cross the boundary between useful information and overwhelming clutter. A copier repairman does not need diagrams to replace the most commonly broken belt. In fact, such interference would be considered annoying. In order to assist the user unobtrusively, the wearable computer must model its user's knowledge, actions, goals, and even emotions. This paper has presented systems to track the user's position, visual field, and current interests as revealed by what is being typed. However, a more personal and striking interface may be possible if the user's emotional affect can be sensed as well.
Emotional affect plays a large part in everyday life. In fact, there is evidence that without affect, even rational intellect is impaired [Damasio, 1994]. To date, computer interfaces have mostly ignored human affect. However, wearable computers, which are in contact with their users in many different contexts, allow an unprecedented opportunity for affect sensing. Picard discusses ways in which computers might recognize affect [Picard, 1995], as well as a number of potential applications of affective wearables. To this end, we have begun to interface temperature, blood volume pressure, galvanic skin response, foot pressure, and electromyogram biosensors with our wearable computers. While simply providing a body ``status line'' overlay for the user's benefit is an interesting application, we hope to create a sophisticated model of the user by combining affect and environment sensing as well as pattern recognition techniques similar to [Orwant, 1993].
Through sensor data and the user model, the wearable computer can track the state of its user and adjust its behavior accordingly. For example, suppose the user is attending an important business lunch. The user's computer, realizing that the user does not want to be disturbed, should take messages if phone or electronic messages arrive. However, in the case of an emergency message, the computer should understand enough of the context to grab the user's attention immediately. The computer should also be able to identify urgent or time-critical messages [Schmandt, 1994] and wait for a break in the conversation to post a summary message discreetly onto the user's heads-up display.
A user model should also predict the user's next action or state. Such information can be used to allocate resources pre-emptively. For example, suppose the user enjoys music while he's working. When working on a late night project, the user likes hard rock to keep him alert. However, during the day, the user prefers classical music to lower his stress. Through having learned these preferences, knowing the time of day, and sensing the user's stress level, the wearable computer can predict what the user may want to listen to next and can download potential selections over the wireless network.
This predictive ability becomes vital when network-stored video clips are to be used for overlays during repair tasks. For example, the wearable computer may begin downloading diagnostic tools for near-future use while displaying an informative overlay for a current repair task.