Academic journal article
By Witmer, Bob G.; Sadowski, Wallace J.
Human Factors , Vol. 40, No. 3
Virtual environments (VEs) and their interfaces typically are designed to recreate some of the sensations and experiences that characterize real-world environments. The success of this endeavor has been limited by both the available technologies and a lack of information regarding what users need in order to perceive and perform in VEs as they do in real environments. Although technologies for providing spatialized audio, force, and tactile feedback and stereoscopic visuals exist, they do not provide an experience that is indistinguishable from reality (Stuart, 1996). In fact, VE experiences are often so different from real-world experiences that even brief VE exposures can produce eyestrain, nausea, and disorientation (Bailey & Witmer, 1994). To examine the behavioral consequences of differences between VEs and real-world environments, controlled experiments comparing performance in the two environments are required. It is important to link the impairment of VE performance to the technological deficiencies responsible for that impairment. The degree to which transfer of training is affected is also a critical issue.
The ability to judge distance accurately is essential to many real-world tasks, including navigation, aiming, and shooting. Direct comparisons of verbal distance estimates in VEs and real environments suggest that observers are less accurate in estimating distance in VEs than in the real world (Lampton, Singer, McDonald, & Bliss, 1995; Witmer & Kline, 1997). Using viewing distances from 3 to 33 m, Witmer and Kline (1997) found that real-world estimates averaged about 75% of the true distance, whereas VE estimates averaged only 50%. Wright (1995) reported that underestimates in VEs ranged from 41% to 72% of the true distances, whereas real-world estimates generally averaged 87% to 91% of the true distances. The relatively poor VE performances reported are probably not the result of using low-end VE display devices, as two of the three experiments employed wide field of view (FOV), high-resolution display devices.
In real-world settings, research has shown that distance judgments can be quite accurate when participants are asked to first view an object and then walk to it without further visual guidance (Elliott, 1987; Laurent & Thomson, 1988; Rieser, Ashmead, Talor, & Youngquist, 1990; Steenhuis & Goodale, 1988; Thomson, 1983). This nonvisually guided locomotion (NVGL) procedure yielded relative errors that were 2% to 8% of the true distances at viewing distances up to 22 m. Rieser et al. (1990) concluded that accurate walking without vision to previously seen targets indicates that efferent or proprioceptive information about locomotion is closely calibrated to visually perceived distance. Unlike verbal distance estimates, which depend on one's concept of what constitutes a foot or a meter, NVGL appears to provide an accurate, unbiased measure of perceived distance. If an environment (e.g., a VE) distorts perceived distance, however, then NVGL should accurately reflect those distortions.
Thomson (1983) conducted the prototypical study of NVGL. Participants viewed a target for 5 s and then, with their eyes closed and their hearing masked, walked distances of 3 to 21 m, stopping when they believed they had reached the target. Distance judgments were within 24 cm of the target for 84% of all walks up to the 9-m mark. However, errors increased sharply to 150 cm or more between 12 m and 18 m. Thomson attributed the sharp drop in performance to the inability to hold spatial knowledge in short-term memory long enough to cover the longer distances. To test this notion, Thomson introduced variable delays between visual obscuration and beginning to walk, finding support for his hypothesis. However, several replication attempts (Elliott, 1986, 1987; Rieser et al., 1990; Steenhuis & Goodale, 1988) did not support Thomson's hypothesis, but all found that variable error increased linearly with distance. …