This study proposes a new methodology for conducting synchronous remote usability studies using a three-dimensional virtual usability testing laboratory built using the Open Wonderland toolkit. This virtual laboratory method is then compared with two other commonly used synchronous usability test methods: the traditional lab approach and WebEx, a web-based conferencing and screen sharing approach. A study was conducted with 48 participants in total, 36 test participants and 12 test facilitators. The test participants completed 5 tasks on a simulated e-commerce website. The three methodologies were compared with respect to the following dependent variables: the time taken to complete the tasks; the usability defects identified; the severity of these usability defects; and the subjective ratings from NASA-TLX, presence and post-test subjective questionnaires. The three methodologies agreed closely in terms of the total number defects identified, number of high severity defects identified and the time taken to complete the tasks. However, there was a significant difference in the workload experienced by the test participants and facilitators, with the traditional lab condition imposing the least and the virtual lab and the WebEx conditions imposing similar levels. It was also found that the test participants experienced greater involvement and a more immersive experience in the virtual world condition than the WebEx condition. These ratings were not significantly different from those in the traditional lab condition. The results of this study suggest that participants were productive and enjoyed the virtual lab condition, indicating the potential of a virtual world based approach as an alternative to the conventional approaches for synchronous usability testing.

Aircraft inspection is a vital element in assuring safety and reliability of the air transportation system. The human inspector performing visual inspection of an aircraft is the backbone of this process and training is an effective strategy for improving their inspection performance. Previous studies have shown offline feedback training to be effective in improving subsequent visual inspection performance. Because experienced inspectors are known to adopt a better inspection strategy than novices, providing visualization of experts' cognitive processes a priori can accelerate novices' adoption of the experts' strategy. Using eye tracking equipment, we record the point of regard of an expert inspector performing an inspection task in a virtual reality simulator. Analysis of their eye movements leads to a visualization of their scanpaths and allows us to display the inspector's visual search (hence cognitive) strategy. We show how providing this type of scanpath-based feedforward training of novices leads to improved accuracy performance in the simulator coupled with an observed speed-accuracy tradeoff. We contend that the tradeoff results from trained novices adopting a slower paced strategy through increased fixation durations, suggesting trained novices learn a more deliberate target search/discrimination strategy that requires more time to execute.

This paper evaluates the use of Visual Deictic Reference (VDR) in Collaborative Virtual Environments (CVEs). A simple CVE capable of hosting two (or more) participants simultaneously immersed in the same virtual environment is used as the testbed. One participant's VDR, obtained by tracking the participant's gaze, is projected to co-participants' environments in real-time as a colored lightspot. We compare the VDR lightspot when it is eye-slaved to when it is head-slaved and show that an eye-slaved VDR helps disambiguate the deictic point of reference, especially during conditions when the user's line of sight is decoupled from their head direction.

As the sophistication of computer technology increases, the number of systems in which both a human and a computer take controlling roles will increase. Effective communication between the human and computer is essential to prevent redundant or interfering actions. If a method for accurately predicting the human's actions can be determined, the computer can use the predictions to complement the human's actions rather than interfere with them. This correspondence develops and empirically validates a two-stage mathematical model of human performance in a situation with real time, multiple tasks, and process control. The model is a consistent and accurate predictor of human performance, correctly predicting over 80 percent of the subjects' actions over a range of situations. A subsequent study is suggested to assess the value of this model when the computer is implemented as a parallel decisionmaker, selecting its own actions to complement rather than conflict with the human decisionmaker.

When tasks are allocated dynamically within a human-computer system, system performance depends upon effective communication between the human and the computer. The results are presented of simulation studies which investigate two means of human-to-computer communication: implicit communication, in which the human's planned actions are conveyed to the computer through a model of the human's action strategy; and explicit communication, in which the human overtly transmits decisions to the computer. The results indicate that the effectiveness of implicit communication depends upon both the predictive validity of the model employed and the computer task allocation strategy built upon the model. The effectiveness of explicit communication depends upon the time the human must devote to transmitting action plans to the computer. These simulations indicate the payoffs that can be realized by optimization of implicit and explicit human-computer communication. They also provide an efficient means to determine the appropriate mix of these communication techniques for a given situation. Perhaps the most interesting finding is that implicit communication based upon a model possessing only modest predictive validity is capable of enhancing system performance. This suggests that implementation of implicit communication within human-computer systems need not require the development and use of complex all-encompassing models of human performance.

When a human and computer perform similar tasks in parallel, it is important that there be effective communication between the two entities. Because explicit communication may add to the human's workload, an implicit method of communication is suggested in which the computer references a model of human performance to predict the human's actions. The computer then determines its own actions so as to complement rather than conflict with the predicted human actions. A two-stage mathematical model of human performance is employed in an experimental situation in which both a human and computer act as decisionmakers. Implementation of the model significantly improves the human's performance, without degrading the computer's performance. Research into additional experimental and real-world situations is suggested.

When a human and computer perform similar tasks in parallel, it is important that an effective line of communication exist between the two entities. Since overt communication may add to the human's workload, an implicit method of communication is suggested in which the computer has a model of human performance on which to base actions. A two-stage model of human performance is employed in an experimental situation in which both a human and a computer act as decision makers. Results indicate that the implementation of a model significantly improves the human's performance and the overall system performance, without degrading the computer's performance. Research into additional experimental and real-world situations is suggested.

It is proposed that human decisionmaking performance in multiple process monitoring situations can be modeled in terms of the detection of process related events and the allocation of attention among processes once events are felt to have occurred. An elementary pattern recognition technique, discriminant analysis, is used to generate estimates of event occurrence probability. A queueing theory framework is then utilized to incorporate these probabilities as well as other task characteristics into the solution of the attention allocation problem. The performance of the model is compared with that of subjects in two experiments.