Training, transfer, and retention of three-dimensional ... - IOS Press

223

Journal of Vestibular Research 12 (2002/2003) 223–238 IOS Press

Training, transfer, and retention of three-dimensional spatial memory in virtual environments Jason T. Richards a,∗, Charles M. Omana, Wayne L. Shebilske b, Andrew C. Bealla,∗∗ , Andrew Liua and Alan Natapoffa a

b

Man Vehicle Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139, USA Department of Psychology, Wright State University, Dayton, OH 45435-0001, USA

Received 6 June 2002 Accepted 22 November 2002

Abstract. Human orientation requires one to remember and visualize spatial arrangements of landmarks from different perspectives. Astronauts have reported difficulties remembering relationships between environmental landmarks when imagined in arbitrary 3D orientations. The present study investigated the effects of strategy training on humans’ 1) ability to infer their orientation from landmarks presented ahead and below, 2) performance when subsequently learning a different array, and 3) retention of configurational knowledge over time. On the first experiment day, 24 subjects were tested in a virtual cubic chamber in which a picture of an animal was drawn on each wall. Through trial-by-trial exposures, they had to memorize the spatial relationships among the six pictures around them and learn to predict the direction to a specific picture when facing any view direction, and in any roll orientation. Half of the subjects (“strategy group”) were taught methods for remembering picture groupings, while the remainder received no such training (“control group”). After learning one picture array, the procedure was repeated in a second. Accuracy (% correct) and response time learning curves were measured. Performance for the second array and configurational memory of both arrays were also retested 1, 7, and 30 days later. Results showed that subjects “learned how to learn” this generic 3D spatial memory task regardless of their relative orientation to the environment, that ability and configurational knowledge was retained for at least a month, that figure rotation ability and field independence correlate with performance, and that teaching subjects specific strategies in advance significantly improves performance. Training astronauts to perform a similar generic 3D spatial memory task, and suggesting strategies in advance, may help them orient in three dimensions. Keywords: Vision, vestibular, spatial orientation, spatial memory, mental imagery, mental rotation, training

1. Introduction Human orientation requires one to remember spatial arrangements of visual landmarks and visualize them ∗ Corresponding author: Jason T. Richards, M.S., Wyle Laboratories, Inc., 1290 Hercules Drive, Suite 120. NL/266, Houston, TX 77058, USA. Tel.: +1 281 483 3730; Fax: +1 281 244 5734; E-mail: [email protected]. ∗∗ Current Address for co-author: Andrew C. Bell, Department of Psychology, University of California, Santa Barbara, CA 93106, USA.

from different perspectives. On Earth, this cognitive process is effortless and reliable, perhaps because the only large mental rotations required are usually about a single axis aligned with gravity. However, astronauts floating in arbitrary orientations have reported difficulties remembering relationships between landmarks inside their spacecraft, and difficulties developing a mental model of the space station structure (Richards, et al., this issue). These difficulties have lead to errors when navigating between modules or when reaching for tools. Such spatial memory problems could jeopardize mission objectives and crew safety in some circum-

ISSN 0957-4271/02/03/$8.00  2002/2003 – IOS Press. All rights reserved

224

J.T. Richards et al. / Training, transfer, and retention of three-dimensional spatial memory in virtual environments

stances. The objectives of the current experiment are to investigate the ability of humans to learn to perform three-dimensional spatial memory tasks analogous to those confronted by astronauts, to determine how long ability and configurational knowledge of simple environments can be retained, and to assess whether generic strategy training facilitates the learning process. Humans are usually able to integrate available sensory information to keep track of their orientation in terrestrial environments with relative ease – even when vision is compromised [9,10,12–14,18,21,22]. Despite the ease with which spatial updating is normally achieved, people sometimes lose their sense of direction. When they see an environment from a familiar perspective they re-orient, or “get their bearings.” The ability to mentally visualize the spatial layout of an environment from different viewpoints is presumably advantageous. Observers of a previously unfamiliar environment tend to remember the spatial structure of it from one or more “prototypic” viewpoints [8, 28]. Once a prototypic spatial model of an environment has been established, mental transformations are applied to imagine novel perspectives relative to the environment. Subjects generally benefit from imagining self-orientation changes within a stationary environment rather than conversely (i.e., they perform egocentric coordinate transformation within an exostationary coordinate frame). Spatial information is organized and processed in a hierarchical and categorical way into spatial frameworks [23]. Prototypic spatial frameworks are often defined relative to referent directions (e.g., ahead/behind, left/right, above/below) aligned with the subject’s body, but can also be remembered relative to environmentally fixed landmarks. When living on Earth, most large body rotations naturally occur about the body’s head/foot axis. Thus, when visualizing the environment from a new viewpoint, only a single mental transformation of the spatial framework about an axis aligned with gravity is required. What are the limits of human ability to imagine rotations and to reorient in environments when visualization requires mental rotations about an arbitrary axis? Are there spatial orientation illusions that interfere with this process? Skylab, Mir, and Shuttle astronauts have reported momentary disorientation when they float into a module in an unfamiliar orientation inside their spacecraft. Since gravity is absent, there is a natural tendency to perceive whichever interior surface is closest to being beneath their feet as a “floor”. As a result, the floors, walls and ceilings exchange subjective identities. The sudden change in perceived orientation,

termed a “visual reorientation illusion” (VRI) triggers attacks of space sickness [15], or makes crewmembers reach the wrong way for remembered objects or look the wrong way in search of visual landmarks. US astronauts living on the Russian MIR space station and International Space Station (ISS) have reported increased spatial orientation difficulties stemming from the complex architecture of the structures. These large spacecraft consist of multiple modules connected together by one or more “nodes.” The Mir station consisted of a single central node. The ISS may eventually have as many as six. An astronaut floating inside a node is surrounded by six different hatches, each one 90 degrees apart and facing a principal spatial direction. When fully configured, each hatch leads to an adjacent module. The “floors” of some of the modules on Mir and ISS are oriented 90 or 180 degrees apart. Many ISS modules have equipment racks on the ceiling and floor as well as the walls, creating dual visual verticals. As a result, crewmembers say they frequently experience VRIs and lose their sense of direction. Crewmembers apparently rely on hatch labeling and use memorized landmarks and routes when orienting inside station. However, the ability to transform a 3D mental model of the station is arguably essential in certain emergency situations, for example should visual landmarks be obscured due to smoke or darkness. Some studies have shown the potential of virtual environments (VE) for teaching terrestrial spatial tasks [19] and environmental familiarization [11]. Practical virtual reality systems are imperfect due to significant problems caused by sensory cue mismatches. Spatial learning in VE, however, has been shown to benefit performance in emergency rescue missions by allowing subjects to develop a mental model of the environment prior to real-life exposure [25,27]. Astronauts train for flight in 1-G simulators, but usually train in an upright gravitational orientation, leaving little opportunity to develop a 3D mental model of the entire space station by direct experience before they fly. Astronauts have long recognized that it is important to train for spacewalks in ways that give them visual experience working in a variety of different body orientations relative to the spacecraft. Crews routinely rehearse spacewalks in neutral buoyancy facilities and use immersive virtual reality simulators. However, ground simulator modules and nodes cannot be connected in the actual flight configuration. Unfortunately, crews routinely have no opportunity to view or practice visualizing the module interiors from non-upright perspectives. Parker et al. [17] reported that allowing Shuttle


astronauts to practice moving around inside a simulated Spacelab module interior reduced their susceptibility to space sickness. However, the effectiveness of the training on spatial orientation ability within the module was not measured, and the technique is not in routine use. NASA’s National Space Biomedical Research Institute (NSBRI) recently began to develop 3D spatial orientation assessment methods appropriate for astronaut preflight visual orientation training. As part of that effort, our first series of experiments [16,24] studied how quickly and well subjects inside a 6-walled node chamber or a virtual equivalent could learn to predict the relative direction of specific objects in a target array after being instructed to imagine themselves in a specific roll orientation (including upside down) and/or viewing direction relative to a canonical reference frame. Subjects were able to achieve relatively high accuracy within 20 exposures from a given viewpoint, regardless of roll orientation, and learned to orient from a second viewpoint with equal or greater ease. Subjects were free to use whatever strategy they wanted. After the experiment, they said they used a combination of memorization strategies, such as memorizing opposite pairs and corner configurations. Some subjects also reported being able to mentally visualize the objects around them from imagined orientations without the help of any mental rules (i.e., as if they could see the object arrangement in their “mind’s eye”). Manipulation of the subject’s body posture with respect to gravity during our experiment produced a statistically significant but quantitatively small effect on accuracy and response time. This finding encourages us to think that the results of our experiments – necessarily conducted in 1-G – probably apply in 0-G as well. When entering a 6-sided space station node, disoriented astronauts arguably must perform a task slightly different from that we used in our first experiments – that is, rather than being told their orientation, they must infer it from the location of objects in the surrounding visual array. Their ability to do so presumably depends on how well they remember the spatial arrangement of the modules relative to one another (i.e., configurational knowledge) and how well they imagine the arrangement from different perspectives (i.e., mentally rotate the configuration). The objective of the current experiment was to study whether human subjects in a 1-G virtual environment can learn to do an analogous task regardless of their relative orientation to the array, whether experience doing the task in a first array accelerates the process of learning a second array,

225

and how many days their configurational knowledge of one of the arrays and their ability to do the task was retained. We also wanted to see whether it was helpful to suggest specific strategies – such as memorizing opposite pairs of objects – and whether an individual subject’s success in learning the task correlated with independent measures of field dependence and two- and three-dimensional mental rotation ability.

2. Method Twenty-four students and faculty (ages 18–47) from Massachusetts Institute of Technology participated. Subjects wore a head mounted display (Kaiser Proview80, 65 deg. × 48 deg. field of view per eye, 640 × 480 × 3 resolution, 100% binocular overlap) and an acoustical-inertial head tracking system (IS600 Mark 2, Intersense, Inc., Burlington, MA). Visual scenes were rendered in color stereo by a graphics accelerator equipped PC workstation (25 Hz average scene update rate, 65 msec. lag from head movement to scene motion) using Python/OpenGL/VRUT software. Subjects sat erect for the entire experiment but were encouraged to look around. Testing was conducted for several hours on four days over a 30-day interval (Fig. 1). On the first test day, referred to as the “training day,” we tested subjects’ ability to learn to orient themselves relative to two different three-dimensional picture arrays, one at a time. We retested subjects’ retention of configurational knowledge and skills one, seven and thirty days later. These were referred to as “retention test days” 1, 7, and 30. Before beginning the experiment, all subjects read scripted instructions presented as self-paced slides on a laptop computer that explained the general task and included practice with the trial sequence and keyboard inputs. Example tasks were performed inside a 3D practice environment similar to the experimental environments (Fig. 2). The instructions also encouraged the use of visualization whenever possible. All subjects were given the same 5 practice trials in the practice environment to become familiar with the procedures. Twelve subjects (strategy group) received instructions that taught specific strategies for remembering the locations of pictures relative to each other. Strategies were based on techniques suggested by subjects in our previous experiments [16]. These included: 1) a “baseline orientation strategy” that included remembering pictures as seen from a prototypic viewpoint established during the first 5 trials, 2) an “opposite pairs strategy”

226

J.T. Richards et al. / Training, transfer, and retention of three-dimensional spatial memory in virtual environments Table 1 Group Scores on Paper and Pencil tests Group

Strategy Control

Card Rotation Median Range 121.5 82.0 114.5 110.0

Test Cube Comparison Median Range 24.0 34.0 25.0 36.0

GEFT Median Range 18.5 16.0 21.0 13.0

Fig. 1. Overall experimental timeline. Exit interviews were conducted at the end of each experimental session (indicated by vertical dashed lines).

consisting of remembering which pictures were located on walls directly across from each other, and 3) a “triad strategy” that required remembering three pictures that define a corner. The remaining 12 subjects (control group) received an identical set of instructions but excluding the strategy lessons, thereby requiring them to develop their own strategy for remembering the picture arrangement. Our previous experiments showed a correlation between performance on a related spatial memory task and conventional paper-and-pencil tests of visual field dependence (Group Embedded Figures Test, or GEFT) [26] and two- and three-dimensional mental rotation ability (Card Rotations and Cube Comparisons Tests) [6], so we repeated those tests on the present groups. The strategy and control groups were approximately balanced by the median card rotation and cube comparison test scores (see Table 1). Groups were balanced within 6% for Card Rotations, 4% for Cube Comparisons, and 12% for GEFT scores. During experimental trials, subjects saw two arrays of 2D pictures as shown in Fig. 2. Each picture consisted of four identical drawings of familiar animals rotated by multiples of 90 degrees and symmetrically clustered so the aggregate could be easily recognized from any orientation. The symmetrical cluster prevented subjects from developing ad-hoc pointer rules that would help them locate one object using another; for example, “the turtle’s nose always points toward the fish.” The entire array of pictures could be rotated as an ensemble by computer, so that any picture could be presented on any wall, but the spatial relationships between pictures always remained constant relative to one another. Subjects were told to interpret the rotation as a change in their body orientation with respect to the array.

The successive views seen by the subject during each 19-sec experimental trial on the training day are shown schematically in Fig. 3. First, one picture – the target– was displayed for 2 seconds. Next, only the pictures ahead and below their bodies were revealed for 3 seconds, followed by an empty array displayed for 7 seconds. Subjects had to infer their simulated orientation (i.e., viewing direction and roll orientation) from the identities of the two pictures presented ahead and below. They then had to discern the relative direction to the specified unseen target picture in body coordinates by pushing one of four buttons on a keypad. The buttons were arranged in an upside-down “T” formation, with the center button denoting “behind the back,” and the three adjacent buttons denoting “left,” “right,” and “above the head”. The keypad was mounted in the subject’s transverse plane and none reported any difficulty remembering or using the button mapping. Finally, the complete picture array was revealed for 7 seconds as it would appear from the tested orientation. This allowed the subject to review the relative arrangement of the pictures and rehearse strategies in preparation for upcoming trials. To allow the subjects to learn at least a few spatial relationships between pictures without having to perform mental rotations, the first 4 trials were presented in a constant, prototypic orientation. For the next eight trials, the subject faced a second viewing direction in all roll orientations (i.e., to show all possible “below” pictures) twice each. These 12 trials were considered training trials and were excluded from the analysis. Performance was then studied during the next 24 trials during which each of the 24 possible orientations relative to the picture array was randomly presented and analyzed in successive groups of 8 trials, which were referred to


227

Fig. 2. Wide-angle views of the three picture arrays, as seen from the prototypic orientation. Array A included pictures of frogs (above), bluebirds (below), snakes (left), deer (right), elephants (ahead), and roosters (behind). Array B included pictures of turtles (above), lions (below), parrots (left), trout (right), butterflies (ahead), and giraffes (behind). The Practice Array included pictures of a star (above), a pound sign (below), hearts (left), diamonds (right), spades (ahead), and clubs (behind).

Fig. 3. Schematic of 3D Spatial Learning Experiment timeline for each trial. Subject was shown target picture, pictures ahead and below, then during “Memory” phase had 7 more seconds to indicate relative target direction in body axes. Finally, during “Study” phase, all 6 pictures are shown. Subject found target and then studied array for the remaining time available.

as “sets”. We wanted to verify that subjects could learn to perform the mental transformations appropriate for any orientation. In our previous study [16] subjects had trained in only two of the six possible viewing directions. On theoretical grounds [1] we expected that random rather than blocked sequencing of viewing direction would ultimately enhance performance. Hence viewing direction as well as roll orientation were randomized and balanced. Only one of four possible relative target directions was presented for each combination of viewing direction and roll orientation to minimize experiment time and fatigue. Each of the four target directions (left, right, above and behind) relative to the subject’s body was presented once every four trials. Following a short break, the subject repeated an identical sequence of 12 practice trials and 3 sets of 8 experimental trials in a second, different environment. Within a set, trials were presented in a pseudo-random fashion, counterbalanced for order across roll orientation and relative target direction. Data were recorded

over the last 3 sets for the first array and the last three sets for the second array, denoted 1–3 and 4–6, respectively, in Fig. 4 below. The order of presentation of the picture arrays was reversed for half the subjects to control for intrinsic differences in “learnability” between the two arrays. Average response time (RT) and percent correct (%C) were calculated from the inferred target location push button data for each eight-trial set. If the subject failed to respond, an incorrect response and maximum RT were recorded as data for that trial. RT was measured from the moment the ahead and behind pictures appeared. Subjects were instructed to respond accurately, but as quickly as possible. We considered %C as a direct measure of accuracy of spatial memory and RT as an indirect measure of spatial task difficulty. Since subjects usually took more time to respond when they were uncertain of their answer, we expected the two measures to be approximately inversely correlated. Based on results from previous studies [2–4] and expected differences in spatial memory strategies used to

228


Fig. 4. Mean performance for all targets, by set, within group on the training day (upper panel: %C, lower panel: RT; error bars are + 1 SEM).

Fig. 5. Mean performance for above/behind targets, by set, within group on the training day (upper panel: %C, lower panel: RT; error bars are + 1 SEM).

find them, we resolved the four possible relative target directions into two categories for the purpose of analysis: above/behind and left/right. Subjects returned on retention test days 1, 7, and 30 to complete two separate tasks that measured the retention of spatial skills. In the Configurational Knowledge Test (CKT), subjects were instructed to reconstruct the spatial relationships of the pictures relative to one another from memory alone and as they appeared from the prototypic viewpoint. The subject chose pictures from a palette and placed them onto the appropriate wall of the virtual chamber using keyboard controls. Each picture was displayed on the chosen wall upon placement. The time required for subjects to complete the CKT was recorded. Subjects were free to take as much time as they needed and could revise their choices if desired. Responses were scored on an ad hoc scale based on the geometric complexity of the difference between the subject’s indication and the prototypic viewpoint. Scores were assigned to responses according to whether the errors involved simple rotations and inversions of pictures in space relative to their respective locations in the prototypic viewpoint, or whether more complex transformations of the arrangement occurred. A score

of 0 was assigned to responses that perfectly matched the configuration as seen from the prototypic viewpoint. Scores 1–9 were assigned to responses that were rotated in only the roll (1–3), yaw (4–6), or pitch (7–9) plane relative to the prototypic viewpoint. Scores 10– 12 were assigned to responses that included an inversion of two pictures in only the left-right, up-down, or front-behind directions, respectively. Other responses involving unique combinations of rotations in multiple planes, combinations of rotations and inversions, inversions of two pairs of pictures, and inversions of all three pairs of pictures were assigned scores 13–129. Responses in which subjects exchanged pictures that were located on adjacent surfaces were not included in our scale and were assigned an arbitrary score of 130. Overall, a low CKT score indicated relatively accurate configurational knowledge, and a high score indicated that the subject confused the locations of many of the pictures. The CKT was completed for both picture arrays in the order in which they were learned on the training day. In the second task, called the Spatial Orientation Test (SOT), subjects completed trials identical to the last 24 trials completed in the second picture array on the train-


229

ing day except without visual feedback for correct picture locations (see Fig. 3 above). The 24-trial sequence of all possible orientations was repeated twice for a total of 48 SOT trials. Before beginning the SOT, subjects were allowed to study the correct arrangement of the second picture array for 20 seconds even if their CKT response was erroneous. This prevented poor configurational knowledge from confounding evaluation of spatial orientation skills. On all test days after completing the tasks, subjects were asked standardized multiple-part, open-ended questions regarding their learning strategies, difficulty in going from one picture array to the next, the relative differences in learning/recalling the two different picture arrays, and their perception of the effect of time between tests. Responses were retrospectively categorized and tabulated. Statistical analysis was performed using conventional software packages (Systat v9.0, SPSS, Inc.; StatXact-4, Cytel, Inc.). Analyses for training day and SOT trials were performed by relative target direction catogary. The experimental session on the training day lasted approximately 2 hours. Experimental sessions on retention test days 1, 7, and 30 each lasted a total of approximately 30 minutes.

3. Results 3.1. Learning effects on the training day We evaluated %C and RT performance on the training day for all (Fig. 4), above/behind (Fig. 5), and left/right (Fig. 6) relative-target-directions, by group. We were only interested in results for trials in which subjects were challenged with all viewing directions and roll orientations in each picture array (first picture array: sets 1–3; second picture array: sets 4–6). The 12 initial trials – 4 upright and 8 facing a second viewing direction – of each array were for training and not included in the analyses. In the figures, the vertical line separates results for the first and second picture arrays. By the end of set 6, overall %C had risen to the 75– 80% range for both groups, well over chance (%25). For left/right targets, %C performance was similar between groups, reaching only about 60% by the end of either picture array, but steadily increasing over sets. For above/behind targets, %C for the control group dropped from 85% to 65% when the picture array changed, but rose to previous levels by the end of the second picture array. Strategy group %C performance for above/behind targets was unaffected by the change

Fig. 6. Mean performance for left/right targets, by set, within group on the training day (upper panel: %C, lower panel: RT; error bars are + 1 SEM).

in picture array, remaining at about 95% correct, and reached 100% by the end of the second picture array. RT performance of the strategy group was generally faster for above/behind targets and slower for left/right targets when compared to that of the control group. To examine the improvement in %C for all targets observed on the training day, we analyzed each group’s performance separately using nonparametric Friedman ANOVA across sets by picture array. Improvement in RT performance was analyzed using repeated measures ANOVA with group, picture array, and set as factors. The improvement in %C between sets for the control group was significant in both the first and second picture arrays ({first, second}: χ 2 = {9.80, 6.78}, df = 2, exact p < {0.01, 0.05}). No significant improvement was found in %C performance between sets in either picture array for the strategy group (p > 0.05). However, a Page test [5], whose statistic is denoted “pa(x),” across sets showed a significant increasing trend in %C performance in the first picture array for this group [pa(x) = 1.83, df = 2, exact p < 0.05]. The improvement in RT performance was significant with main effects of picture array and set [F (1, 21) = 16.87,

230

J.T. Richards et al. / Training, transfer, and retention of three-dimensional spatial memory in virtual environments Table 2 Post-Hoc Tabulation of Exit Interview Responses on the Training Day Questions: 1. Do you think your ability to perform the task improved in the second training environment because of the experience in the first? 2. What strategies did you use to learn the picture configurations?

3. Were you ever able to mentally visualize the pictures around you without the help of any rules?

Answer Category Strategy (N = 12) Control (N = 12) YES: YES: 11 11 NO: NO: 1 1 Pairs + Visualization: 4 4 Baseline + Pairs + Triads (success- Pairs + Baseline: ful use of right hand rule): 4 5 Baseline + Pairs + Triads (trouble Other with right hand rule): 4 3 NO (dependent on rules): NO (dependent on rules): 8 8 YES: YES 4 4

p < 0.01 and F (2, 42) = 6.35, p < 0.01, respectively]. Group and interaction effects were not significant. Results indicate that all subjects ultimately learned to perform the task with relatively high accuracy in one or both picture arrays. Figures 5 and 6 indicate an apparent difference in subjective difficulty between relativetarget-direction categories. This was verified in exit interviews when subjects claimed to use different strategies for finding above/behind and left/right targets, and it is the reason data is later analyzed by relative-targetdirection category. Interestingly, all who claimed to use visualization also claimed to use the “pairs” strategy (Table 2). Remembering opposite pairs was relatively easy to do, and it allowed subjects to locate two of the hidden pictures, which subsequently may have helped them visualize the picture array around them. The notion that subjects are “learning how to learn” to do the task in the first picture array is supported by the overwhelming majority of subjects (22 out of 24) who responded “yes” to Question 3 of the exit interview and that, despite the change in picture array, the strategy group’s %C performance significantly improved across all sets on the training day. 3.2. Learning how to learn By our definition, “learning how to learn” required that learning occur faster (i.e., faster rise or fall time) in the second picture array than in the first, but not necessarily that a higher level of asymptotic performance be achieved. To test whether subjects learned how to learn to do the three-dimensional spatial orientation task, we compared %C and RT performance between picture ar-

rays, by early (sets 1 and 4), middle (set 2 and 5) and late (set 3 and 6) sets within groups separately using Friedman and paired t tests, respectively. When all targets were taken together, significant differences in %C performance could not be demonstrated in any set between picture arrays for either group. The strategy group had faster early- and middle-set RT performance over all targets in the second picture array than in the first picture array [{early, middle}: t(11) = {3.1, 2.7}, p < {0.01, 0.05}]. The control group had faster middleset RT performance over all targets in the second picture array than in the first picture array [t(11) = 3.67, p < 0.01]. When targets were taken by relativetarget-direction category, the strategy group had higher middle-set %C performance for above/behind targets in the second picture array than in the first picture array (χ2 = 6.0, df = 1, exact p < 0.05). A corresponding improvement in early- and middle-set RT performance for above/behind targets was also found for this group [{early, middle}: t(11) = {2.9, 3.0}, p < {0.05, 0.05}]. For the control group, only middleset RT performance for above/behind targets improved significantly between the first and second picture arrays [t(11) = 2.3, p < 0.05]. No significant differences in %C performance for above/behind targets between picture arrays was found in any set for the control group. For left/right targets, no improvement in %C performance was found between picture arrays for either group in either measure. Improvement in RT performance for all targets between picture arrays was largely driven by improvement for above/behind targets and was seen earlier for the strategy group than for the control. Improvement was mainly seen in middle-set performance for


231

Table 3 CKT Median Error Scores and Mean RT by Day Group

First Environment Second Environment Day 1 Day 7 Day 30 Day 1 Day 7 Day 30 Error Score RT (sec) Error Score RT (sec) Error Score RT (sec) Error Score RT (sec) Error Score RT (sec) Error Score RT(sec) Trained 0.0 109.4 12.0 70.2 14.5 95.1 0.0 91.7 0.0 61.3 0.0 45.1 Control 115.0 93.5 130.0 83.8 130.0 86.8 4.5 67.7 0.0 51.9 5.0 54.2

above/behind targets, especially for the strategy group. Training and practice trials completed prior to the beginning of the experiment should have eliminated any confounding effects associated with subjects familiarizing themselves with experiment display and controls. We attribute the observed improvements in performance between picture arrays to the fact that subjects learned how to learn to do the task. 3.3. Effect of strategy training on the training day To test whether generic strategy training helped the strategy group’s performance on the training day, we compared %C and RT performance between groups by picture array, set, and relative-target-direction category using a Kruskal-Wallis test. For above/behind targets, the strategy group had higher early- and middle-set %C performance in the second picture array than the control [set {4, 5}: χ2 = {5.3, 9.1}, exact p < {0.05, 0.01}]. A corresponding difference between groups was found for middle-set RT performance in the second picture array (set 5: χ2 = 7.0, exact p < 0.01). In contrast, the control group had significantly faster RT performance for left/right targets than the strategy group in the last set of both picture arrays [set {3, 6}: χ 2 = {4.1, 4.3}, exact p < {0.05, 0.05}]. We could not demonstrate a significant difference between groups in %C performance for left/right targets. Strategy subjects clearly benefited from the “pairs” strategy, responding faster and more accurately for above/behind targets by the early set of the second picture array than the control group at any other point in the experiment. It was thought that the “Triad” strategy (remembering and learning to visualize the surfaces that intersect at a corner) would help subjects organize spatial relationships in visual memory. The example strategy we presented to them during training, in retrospect, was evidently cumbersome. To help subjects visualize the pictures around them, the strategy instructions suggested that subjects could assign the index finger, the middle finger, and the thumb to the pictures located above, ahead, and to the left, respectively, in the prototypic orientation and imagine orthogonally arranging the fingers such that each pointed to its assigned direc-

tion. For trials when visualization was difficult, subjects could determine their simulated orientation relative to the prototypic one by physically rotating the orthogonal finger arrangement to match the locations of their assigned pictures. We suspect that, instead of trying to use visualization exclusively, most subjects obediently followed this hand-based strategy and actually performed hand motions associated with our example. All subjects found it time-consuming whether they thought it helpful or not. This could explain why the strategy group responded slowly for the left/right targets. It could also explain why the strategy group did not respond more accurately for left/right targets.

3.4. Effect of visualization ability In the exit interview on the training day, four strategy and four control subjects (n = 8) responded “yes” to the question “Were you ever able to mentally visualize the pictures around you without the help of any rules? i.e. Could you ‘see’ the pictures in your ‘mind’s eye’?” (see Table 2 above), herein called the “visualization group”. The remaining subjects (n = 16), the “non-visualization group”, responded “no” to this question. To assess whether visualization was associated with superior performance on the training day, a post-hoc comparison between groups was performed using Kruskal-Wallis tests by set and relative target direction. The visualization group had significantly higher %C for left/right targets than the non-visualization group in the last set of the first picture array and in all three sets of the second picture array [set {3, 4, 5, and 6}: χ 2 = {4.0, 4.5, 5.6, 8.7}, exact p < {0.05, 0.05, 0.05, 0.005}] (Fig. 7). No significant differences in performance for above/behind targets were found between groups. The visualization group also had significantly higher scores on all three paper-andpencil tests than the non-visualization [{Card, Cube, and GEFT}: χ2 = {5.7, 8.2, 4.0}, p(Monte Carlo

232


3.5. Effect of time on configurational knowledge

Fig. 7. Mean %C for left/right targets, by set and visualization group on the training day (error bars are + 1 SEM).

Control Strategy

Fig. 8. Mean %C for above/behind targets for non-visualization subjects, by set and training group (error bars are + 1 SEM).

estimate)1 < {0.05, 0.005, 0.05}]. We also compared the performances of the strategy and control subjects within these two groups separately using KruskalWallis tests by set and relative-target-direction category on the training day. For the visualization group, there were no significant differences between strategy and control subjects’ performances for either relative target direction for any set. Among non-visualization subjects, as shown in Figure 8, those who were strategy trained had better early and middle set %C for above/behind targets than the control in both picture arrays [set {1, 2, 4, 5}: χ 2 = {4.5, 4.5, 5.1, 8.4}, exact p < {0.05, 0.05, 0.05, 0.01}]. No effect of strategy training on performance for left/right targets was found among non-visualization subjects.

1 The StatXact software approximates the true distribution of the calculated statistic, not with the usual asymptotic chi-square, but with a Monte Carlo simulation whose result, marked “p(MC estimate)”, should be closer to the true p-level which, in these applications, it cannot calculate exactly.

To assess subjects’ abilities to retain configurational knowledge over time, we analyzed the change in CKT error scores and RT within groups using the Page Test across retention test days by set and picture array (Table 3). CKT error scores and RT were generally better for the second picture array due to the fact that visual feedback was not given for the first picture array. Both groups’ median CKT error scores for the second picture array were spatially correct (i.e., error score < 10) over all three retention test days. CKT error scores for the strategy group significantly worsened over days for the first picture array [pa(x) = 1.8, df = 2, exact p < 0.05]. No significant change was found in CKT error scores over days for the first array among control subjects, but they had poor CKT error scores (median >= 115) for that array on all three retention test days. The strategy group showed significant RT improvement for the second picture array over days [pa(x) = −2.6, df = 2, exact p < 0.005]. Neither group showed significant RT improvement for the first picture array over days. To assess the effect of strategy training on retention of configurational knowledge over time, CKT error scores were compared between groups by picture array and day. Comparisons by day yielded no significant differences, but when scores were collapsed across days, the strategy group had significantly lower CKT error scores for the first picture array than did the control group (χ 2 = 4.2, df = 1, p < 0.05). Results for the second picture array show that configurational knowledge is better retained when subjects’ memory for the correct configuration is refreshed and spatial orientation practice is given. Without feedback or practice, CKT scores and completion times for the first picture array degraded over time for the strategy group and were consistently bad for the control. 3.6. Effect of time on spatial ability To assess the effect that the layoff between retention test days had on SOT performance, we compared %C and RT on the last set of one day with the first set of the next (i.e., set 6 vs. 7, 12 vs. 13, and 18 vs. 19) using Kruskal-Wallis and paired-t tests as before. All targets were taken together since no significant differences in %C or RT were found between days for either relative target direction category separately (Fig. 9). The increase in RT between the end of retention test day 7 and the start of retention test day 30 (i.e., between sets 18 and 19) was significant for the control group


[t(11) = −4.72, p < 0.005]. The decrease in %C seen between the end of the training day and the start of retention test day 1 (i.e. between sets 6 and 7) was significant for the strategy group [χ 2 = 6.40, df = 1, exact p < 0.05]. To assess whether subjects’ abilities to perform SOT tasks changed over the course of the 30-day period, we analyzed trends in %C and RT using Page tests over all sets within each group. The general trend of improvement in %C across all days (sets 7–24) was significant for both groups [{control, strategy}: pa(x) = {3.3, 3.3}, p(Monte-Carlo estimate) < {0.0005, 0.0005}]. The strategy group had a corresponding significant trend of improvement in RT over all days [pa(x) = −5.7, p(Monte-Carlo estimate) < 0.0001], but the control group did not. Although responses to exit interviews on retention test days 1, 7, and 30 were not recorded, they suggested that subjects were able to recover spatial abilities over time with relative ease after initial re-familiarization with environments. One subject reported that remembering how to do tasks on retention test days 7 and 30 was “kind of like riding a bike – once you get back on, you remember how to do it almost immediately” after about 3 or 4 trials. The decrease in accuracy seen for the strategy group is probably related to the difficulty they had implementing the “triad” strategy for left/right targets. It may also be the case that visualization skills are more sensitive to time than is the ability to remember and implement strategies. The fact that the 23-day layoff did not affect strategy group performance and the observed improvement in RT performance over days indicates that formal presentation of strategy training lessons may aid skill retention. 3.7. Effect of relative target direction Retrieval of spatial information is apparently aided by the asymmetries inherent in the human body and in gravitational environments: Imaginary objects located on the subject’s left or right are identified slightly (< 0.8 sec) more slowly than those located gravitationally above or below [3,4,7]. Rather than studying the process of learning, the majority of studies on imagined spatial updating have considered the situation in which novel configurations are encountered, or once asymptotic performance has been reached. We wanted to see if there were similar differences in performance between relative-target-direction categories in the present study. To be consistent, we tested for differences in %C and RT performance within each group using a Mann-Whitney test (df = 1 for all

233

tests) between the two relative-target-direction categories in the third and sixth set of every day (i.e. set 3, 6, 9, 12, 15, 18, 21, and 24) where performance was assumed to have neared an asymptotic level (see Fig. 10). The strategy group showed significantly higher %C and shorter RT for above/behind targets than they did for left/right targets in every set tested on all days [sets {3, 6, 9, 12, 15, 18, 21, 24}: χ2 = {10.3, 7.4, 5, 7, 11.2, 8.2, 7.0, 7.0, 7.5}, p < {0.005, 0.01, 0.05, 0.005, 0.005, 0.01, 0.01, 0.01}]. Control subjects showed significantly higher %C for above/behind targets than for left/right targets in set 6 on the training day and in both sets tested on retention test days 1 (sets 9 and 12) and 7 (sets 15 and 18). The control group’s RT was significantly lower for above/behind targets than for left/right targets in both sets tested on retention days 1 and 7 and in the sixth set (set 24) on Day 30. Overall mean differences between relative-target-direction categories were 28% and 3.2 sec for the strategy group and 19% and 1.7 sec for the control group. 3.8. Predictors of task performance We checked for correlations between steady state %C and RT and paper-and-pencil test scores by picture array within group (Table 4). For the control group, steady-state %C in both picture arrays correlated significantly with results of all three paper-and-pencil tests (t test, df = 10), replicating the results found in our previous experiments. For the strategy group, only GEFT scores correlated significantly with steady-state %C and only in the second picture array. Steady-state RT in both picture arrays for the strategy group correlated significantly with the Card rotation test, but not with the GEFT or the Cube comparison test. We also looked for correlations between performance measures and the sex of the subjects. We interpret the results cautiously because only 7 of the 24 subjects were females (4 in strategy, 3 in control). Differences in performance between sexes were analyzed for each day using independent samples t tests and Kruskal-Wallis nonparametric tests. For the control group, males had significantly lower (i.e., better) CKT scores for the first environment on retention day 1 than did females (χ2 = −2.05, p < 0.05). There were no significant differences in performance between sexes for the strategy group. Furthermore, there were no significant differences between sexes for paper-and-pencil test scores.

234


Fig. 9. Mean Performance for all targets by set and day within group. Data for training day (sets 4–6) using second array (upper panel: %C, lower panel: RT; error bars are + 1 SEM).

4. Discussion The exit interview results suggested that control subjects used a combination of self-developed declarative mnemonic rules and mental visualization techniques, as was noted in our first experiment [16]. Our “control” training constituted a form of training in itself: the subjects had a chance to try different strategies for doing the task, and after 24 trials or so, many were successful in finding strategies that worked – especially for above/behind targets. Many subjects memorized opposite pairs. If the target was located on a wall opposite one of the pictures presented ahead or below,

they could infer its relative location immediately using opposite pairs rules rather than visualization. Selfreported use of visualization, however, was statistically associated with superior performance for the left/right targets and predicted by 2/3D figure rotation ability. The baseline orientation and pairs strategies, whether taught or self-discovered, were the key to doing the task. Other mnemonic strategies, such as memorizing the clockwise (or counterclockwise) order in a ring of 4 pictures, were discovered and described by some of the control subjects and may warrant further investigation. Our strategy training was particularly helpful to subjects who did not report the use of visualization.


235

Fig. 10. Mean RT performance for both groups by relative-target direction by day (upper panel: strategy, lower panel: control; error bars are + 1 SEM).

Our training technique could also be improved. The triad strategy appeared to hamper the performance of subjects with worse mental rotation abilities, while the pairs appeared to help their performance. This could account for the lack of a correlation between strategy group performance and paper and pencil tests. Mental imagery was not necessarily learned, but some subjects thought visualization became easier with time. Exit interview results suggested that only 1/3 of our subjects claimed that they used visualization successfully, including as many of the control subjects as strategy subjects, and usually not even in both picture arrays. Despite our emphasis in training on visualization, we probably did not teach anyone to use mental imagery in this relatively short experiment. The benefit

of mental visualization abilities to those who had them was, however, quantitatively demonstrated. We were able to demonstrate an effect similar to that of target location with respect to major planes of body symmetry shown by Tversky and collaborators [3,4, 7]. Their subjects imagined erect, supine and prone body positions with respect to familiar scenes with clear relationships between objects described by narratives. They studied steady-state performance, rather than the learning process or changes in spatial ability and configurational knowledge over time. In addition to finding longer RT for targets on the left or right of imagined body orientations, we also found lower %C for these targets than for other principal body directions – an effect we were not able to demonstrate in our previous experiment.

236

J.T. Richards et al. / Training, transfer, and retention of three-dimensional spatial memory in virtual environments Table 4 Spearman correlation coefficients for paper and pencil tests

df = 10

GEFT Card Rot. Cube Comp. %C First %C Second RT 1 RT 2 ∗∗∗ p

GEFT Strategy 1 0.26

Card Rot. Control Strategy Control 1 0.49 1 1

Cube Comp. Strategy Control

%C First Envir. Strategy Control

0.75∗∗∗

0.61∗

0.37

0.87∗∗∗

1

1

0.15

0.68∗∗

0.30

0.75∗∗∗

0.39

0.65∗

1

1

0.55∗

0.64∗

0.27

0.86∗∗∗

0.46

0.82∗∗∗

0.14

0.93∗∗∗

0.11 0.29

−0.08 0.04

−0.58∗ −0.65∗

−0.35 −0.44

0.11 0.06

−0.29 −0.44

−0.64∗ −0.27

−0.20 −0.10

%C RT 1 Second Envir. Strategy Control Strategy Control

1

1

−0.02 0.05

−0.20 −0.21

1 0.63∗

1 0.87∗∗∗

< 0.005, ∗∗ p < 0.01, ∗ p < 0.05.

The fact that SOT trials were performed only for the second picture array encountered on the training day prevented us from assessing the effect of time on spatial ability in the first picture array. Furthermore, subjects’ performance for the second picture array on retention test days 7 and 30 was probably better than it would have been if the subjects had completed only 24 SOT trials on retention test days 1 and 7, respectively. Results for sets later than the third set on retention test days 1, 7, and 30 (sets 10–12, 16–18, and 22–24) should be interpreted warily. The best measure of the effect of time on configurational knowledge was probably the CKT results for the first picture array, since feedback was not provided on the correct configuration.

5. Conclusions Our ultimate goal was to demonstrate the effectiveness of virtual reality as a potential training countermeasure useful for teaching generic 3D spatial memory skills to astronauts. With further development, our procedure could serve as a means for combating effects of disorientation and illusions that threaten astronauts’ capacity to carry out missions in microgravity within complex three-dimensional structures. This experiment investigated whether subjects could learn to perform generic spatial orientation tasks analogous to those confronting a crewmember in a space station node and retain their skills over time. We were also interested in knowing if one can “learn how to learn” an environment; i.e., whether computer-based training with generic strategies followed by experience in one environment accelerates learning in a second environment and whether visual imagery is important in such tasks. Our hypothesis was that if subjects were taught

a strategy for performing the spatial task, the practice they had with the task in a first environment would develop spatial skills that could then be applied to other similar environments. We hypothesized that teaching generic strategies would decrease the training time required to learn to perform subsequent 3D spatial orientation tasks and make retrieval of spatial memory easier. If training is generic and retained over periods of several weeks it would not necessarily have to be conducted immediately before flight nor in an environment identical to flight. Experimental results confirmed that subjects were able to learn both picture arrays and achieve relatively high overall target accuracy (75–80%) within 36 trials, the last 24 of which were from arbitrary orientations, in any picture array. It was more difficult to learn and retrieve spatial information associated with symmetric body planes (e.g., left and right) than for asymmetric body planes. Subjects who claimed to use mental visualization were better at finding the more difficult targets and had higher scores on 2 and 3 dimensional figure rotation tests. Overall, teaching specific strategies in advance significantly improved learning of generic 3D spatial orientation skills, especially for those who had difficulty with mental imagery or had poor figure rotation ability. Configurational knowledge and spatial skills were retained for at least 30 days. Subjects clearly “learned how to learn” this generic 3D spatial memory task. Strategy trained subjects showed superior overall performance. Conventional tests of field independence and 2 and 3 dimensional figure rotation ability correlated with both of our task performance measures and could be used to determine individual spatial orientation training needs. Taken together, our two spatial memory experiments demonstrate the ability of humans to perform both


reverse (imagined body orientation) and forward (inferred body orientation) tasks. The forward task encountered in the present experiment involved reorientation in all three dimensions, which is analogous to what astronauts face in a space station node. That some subjects in the first experiments and many members of the control group in the present experiment discovered the pairs strategy on their own shows that practice in this sort of task is beneficial, even without formal strategy training. The observed transfer of learning is encouraging and shows that teaching generic memorization strategies and providing practice with mental imagery can help subjects learn to orient in virtual environments. Our earlier finding that performance was substantially unaffected by body posture, and subject’s comments that the task is one which is “done in your head” encourage us to believe that our results in virtual environments will apply to astronauts in real 0-G environments as well. Head-mounted virtual reality displays are practical for spatial orientation training in 1-G and can serve as a practical countermeasure for problems in 0-G. Future experiments should investigate how generic training helps orientation in 0-G environments and its effectiveness compared to environment-specific training. The obvious advantage of generic training is its significantly reduced cost compared to mission-specific training. The latter requires more programming time, higher-resolution displays, and high-end graphics simulators to produce realistic-looking ISS and Shuttle virtual mockups. In addition to our generic training, sessions with mission-specific environments would probably be helpful. Our current generic virtual spatial orientation procedure requires only one virtual environment with a set of generic landmarks, and about 2 hours of training time. Participants believe that they have learned some useful strategies and understand the difficulties of orienting in a three-dimensional environment.

References [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

Acknowledgments The experimental protocol was reviewed and approved by the MIT Committee on the Use of Humans as Experimental Subjects (COUHES). Supported by NASA Cooperative Agreement NCC9-58 with the National Space Biomedical Research Institute, Houston, TX, USA. Mr. Richards’s current address is 1290 Hercules Drive, Suite 120, Houston, TX 77058. Dr. Beall’s current address is Department of Psychology, University of California, Santa Barbara, CA 93106.

237

[15]

[16]

[17]

R.A. Bjork, Memory and metamemory considerations in the training of human beings, in: Metacognition: Knowing about Knowing, J. Metcalfe and A.P. Shimamura, eds, MIT Press, Cambridge, MA, 1994, pp. 185–205. D.J. Bryant and B. Tversky, Mental representations of perspective and spatial relations from diagrams and models, Journal of Experimental Psychology: Learning, Memory and Cognition 25(1) (1999), 137–156. D.J. Bryant and B. Tversky, Internal and external spatial frameworks for representing described scenes, Journal of Memory and Language 31 (1992), 74–98. D.J. Bryant and W.G. Wright, How body asymmetries determine accessibility in spatial frameworks, The Quarterly Journal of Experimental Psychology 52A(2) (1999), 487–508. W.J. Conover, The Page Test for Ordered Alternatives, Chapter in Practical Nonparametric Statistics, 3rd Edition, 1999, pp. 380, Wiley. R.B. Ekstrom, J.W. French, H.H. Harman and D. Derman, Manual for Kit of Factor-Referenced Cognitive Tests, Princeton, NJ: Educational Testing Service, 1976. N. Franklin and B. Tversky, Searching imagined environments, Journal of Experimental Psychology: General 119(5) (1990), 63–76. J. Huttenlocher and C.C. Presson, The coding and transformation of spatial information, Cognitive Psychology 11 (1979), 375–394. Y.P. Ivanenko, R. Grasso, I. Israel and A. Berthoz, The contribution of otoliths and semicircular canals to the perception of two-dimensional passive whole-body motion in humans, Journal of Physiology 502 (1997), 223–233, London. R.L. Klatzky, J.M. Loomis, A.C. Beall, S.S. Chance and R.G. Golledge, Spatial updating of self-position and orientation during real, imagined, and virtual locomotion, Psychological Science 9 (1998), 293–298. G. Koh, Training spatial knowledge acquisition using virtual environments, Thesis (M. Eng.), Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Sciences, 1997. J.M. Loomis, J.A. Da Silva, N. Fujita and S.S. Fukusima, Visual space perception and visually directed action, Journal of Experimental Psychology: Human Perception and Performance 18 (1992), 906–922. J.M. Loomis, R.L. Klatzky, R.G. Golledge, J.G. Cicinelli, J.W. Pellegrino and P. Fry, Nonvisual navigation by blind and sighted: Assessment of path integration ability, Journal of Experimental Psychology: General 122 (1993), 73–91. M.L. Mittelstaedt and S. Glasauer, Idiothetic navigation in gerbils and humans, Zoologische Jahrbucher Abteilungun fur algemeine Zoologie und Physiologie der Tiere 95 (1991), 427– 435. C.M. Oman, B.K. Lichtenberg, K.E. Money, R.K. McCoy, M.I.T./Canadian vestibular experiments on the Spacelab-1 mission: Space motion sickness: symptoms, stimuli, and predictability, Experimental Brain Research 64 (1986), 316–334. C. Oman, W. Shebilske, J. Richards, T. Tubre, A. Beall and A. Natapoff, Three dimensional spatial memory and learning in real and virtual environments, Journal of Spatial Cognition and Computation (2002). D.E. Parker, D.L. Harm, et al., Self-attitude awareness training: an aid to effective performance in microgravity and virtual environments? Proceedings: Conference on Intelligent

238

[18]

[19]

[20]

[21]

[22]

J.T. Richards et al. / Training, transfer, and retention of three-dimensional spatial memory in virtual environments Computer-Aided Training and Virtual Environment Technology, NASA Johnson Space Center, Houston TX, 1993. H.L. Pick, Jr and J.J. Rieser, Children’s cognitive mapping, in: Spatial Orientation: Development and Physiological Bases, M. Potegal, ed, New York: Academic Press, 1982, pp. 107– 128. J.W. Regian, W.L. Shebilske and J.M. Monk, Virtual reality: An instructional medium for visual spatial tasks, Journal of Communication 42 (1992), 136–149. J.T. Richards, J.B. Clark, C.M. Oman and T.H. Marshburn, Neurovestibular effects of long-duration spaceflight: A summary of MIR-Phase 1 experiences, Proceedings of the Sixth Symposium on the Role of the Vestibular Organs in Space Exploration, Portland, OR, 2002. J.J. Rieser, Access to knowledge of spatial structure at novel points of observation, Journal of Experimental Psychology: Learning, Memory and Cognition 15(6) (1989), 1157–1165. J.J. Rieser, D.A. Guth and E.W. Hill, Sensitivity to perspective structure while walking without vision, Perception 15 (1986), 173–188.

[23]

[24]

[25]

[26]

[27]

[28]

E.K. Sadalla, W.J. Burroughs and L.J. Staplin, Reference points in spatial cognition, Journal of Experimental Psychology: Human Learning and Memory 6(5) (1980), 516–528. W.L. Shebilske, T. Tubre, T. Willis, A. Hanson, C. Oman and J. Richards, Simulating Spatial Memory Challenges Confronting Astronauts, Proceedings of the Annual Meeting of the Human Factors and Ergonomics Society (July 30, 2000). P.N. Wilson, N. Foreman and M. Tlauka, Transfer of spatial information from a virtual to a real environment, Human Factors 39(4) (1997), 526–531. H.A. Witkin, P. Oltman, E. Raskin and S. Karp, Manual for the Embedded Figures Tests, Palo Alto, CA: Consulting Psychologists Press, 1971. B.G. Witmer, J.H. Bailey and B.W. Knerr, Virtual spaces and real world places: Transfer of route knowledge, International Journal of Human-Computer Studies 45 (1996), 413–428. M. Wraga, S.H. Creem and D.R. Proffitt, The influence of spatial reference frames on imagined object and view rotations, Acta Psychologica 102 (1999), 247–264.