A new method for generating trails from a person's movement through a virtual environment (VE) is described. The method is entirely automatic (no user input is needed), and uses string-matching to identify similar sequences of movement and derive the person's primary trail. The method was evaluated in a virtual building, and generated trails that substantially reduced the distance participants traveled when they searched for target objects in the building 5–8 weeks after a set of familiarization sessions. Only a modest amount of data (typically five traversals of the building) was required to generate trails that were both effective and stable, and the method was not affected by the order in which objects were visited. The trail generation method models an environment as a graph and, therefore, may be applied to aiding navigation in the real world and information spaces, as well as VEs.

Three levels of virtual environment (VE) metric are proposed, based on: (1) users' task performance (time taken, distance traveled, and number of errors made), (2) physical behavior (locomotion, looking around, and time and error classification), and (3) decision making (i.e., cognitive) rationale (think aloud, interview, and questionnaire). Examples of the use of these metrics are drawn from a detailed review of research into VE wayfinding. A case study from research into the fidelity that is required for efficient VE wayfinding is presented, showing the unsuitability in some circumstances of common metrics of task performance such as time and distance, and the benefits to be gained by making fine-grained analyses of users' behavior. Taken as a whole, the paper highlights the range of techniques that have been successfully used to evaluate wayfinding and explains in detail how some of these techniques may be applied.

Two experiments investigated participants' ability to search for targets in a cluttered small-scale space. The first experiment was conducted in the real world with two field of view conditions (full vs. restricted), and participants found the task trivial to perform in both. The second experiment used the same search task but was conducted in a desktop virtual environment (VE), and investigated two movement interfaces and two visual scene conditions. Participants restricted to forward only movement performed the search task quicker and more efficiently (visiting fewer targets) than those who used an interface that allowed more flexible movement (forward, backward, left, right, and diagonal). Also, participants using a high fidelity visual scene performed the task significantly quicker and more efficiently than those who used a low fidelity scene. The performance differences among all the conditions decreased with practice, but the performance of the best VE group approached that of the real-world participants. These results indicate the importance of using high fidelity scenes in VEs, and suggest that the use of a simple control system is sufficient for maintaining one's spatial orientation during searching.

Three experiments investigated the effect of implementing low-level aspects of motor control for a collaborative carrying task within a VE interface, leaving participants free to devote their cognitive resources to the higher-level components of the task. In the task, participants collaborated with an autonomous virtual human in an immersive virtual environment (VE) to carry an object along a predefined path. In experiment 1, participants took up to three times longer to perform the task with a conventional VE interface, in which they had to explicitly coordinate their hand and body movements, than with an interface that controlled the low-level tasks of grasping and holding onto the virtual object. Experiments 2 and 3 extended the study to include the task of carrying an object along a path that contained obstacles to movement. By allowing participants' virtual arms to stretch slightly, the interface software was able to take over some aspects of obstacle avoidance (another low-level task), and this led to further significant reductions in the time that participants took to perform the carrying task. Improvements in performance also occurred when participants used a tethered viewpoint to control their movements because they could see their immediate surroundings in the VEs. This latter finding demonstrates the superiority of a tethered view perspective to a conventional, human'seye perspective for this type of task.

A set of rules is presented for the design of interfaces that allow virtual objects to be manipulated in 3D virtual environments (VEs). The rules differ from other interaction techniques because they focus on the problems of manipulating objects in cluttered spaces rather than open spaces. Two experiments are described that were used to evaluate the effect of different interaction rules on participants' performance when they performed a task known as “the piano mover's problem.” This task involved participants in moving a virtual human through parts of a virtual building while simultaneously manipulating a large virtual object that was held in the virtual human's hands, resembling the simulation of manual materials handling in a VE for ergonomic design. Throughout, participants viewed the VE on a large monitor, using an “over-the-shoulder” perspective. In the most cluttered VEs, the time that participants took to complete the task varied by up to 76% with different combinations of rules, thus indicating the need for flexible forms of interaction in such environments.

Imagine walking around a cluttered room but then having little idea of where you have traveled. This frequently happens when people move around small virtual environments (VEs), searching for targets. In three experiments, participants searched small-scale VEs using different movement interfaces, collision response algorithms, and fields of view. Participants' searches were most efficient in terms of distance traveled, time taken, and path followed when the simplest form of movement (view direction) was used in conjunction with a response algorithm that guided (“slipped”) them around obstacles when collisions occurred. Unexpectedly, and in both immersive and desktop VEs, participants often had great difficulty finding the targets, despite the fact that participants could see the whole VE if they stood in one place and turned around. Thus, the trivial real-world task used in the present study highlights a basic problem with current VE systems.

Participants used a helmet-mounted display (HMD) and a desk-top (monitor) display to learn the layouts of two large-scale virtual environments (VEs) through repeated, direct navigational experience. Both VEs were “virtual buildings” containing more than seventy rooms. Participants using the HMD navigated the buildings significantly more quickly and developed a significantly more accurate sense of relative straight-line distance. There was no significant difference between the two types of display in terms of the distance that participants traveled or the mean accuracy of their direction estimates. Behavioral analyses showed that participants took advantage of the natural, head-tracked interface provided by the HMD in ways that included “looking around” more often while traveling through the VEs, and spending less time stationary in the VEs while choosing a direction in which to travel.

Two experiments investigated components of participants' spatial knowledge when they navigated large-scale “virtual buildings” using “desk-top” (i.e., nonimmersive) virtual environments (VEs). Experiment 1 showed that participants could estimate directions with reasonable accuracy when they traveled along paths that contained one or two turns (changes of direction), but participants' estimates were significantly less accurate when the paths contained three turns. In Experiment 2 participants repeatedly navigated two more complex virtual buildings, one with and the other without a compass. The accuracy of participants' route-finding and their direction and relative straight-line distance estimates improved with experience, but there were no significant differences between the two compass conditions. However, participants did develop significantly more accurate spatial knowledge as they became more familiar with navigating VEs in general.