Virtual reality training of lucid dreaming

Metacognitive reflections on one's current state of mind are largely absent during dreaming. Lucid dreaming as the exception to this rule is a rare phenomenon; however, its occurrence can be facilitated through cognitive training. A central idea of respective training strategies is to regularly question one's phenomenal experience: is the currently experienced world real, or just a dream? Here, we tested if such lucid dreaming training can be enhanced with dream-like virtual reality (VR): over the course of four weeks, volunteers underwent lucid dreaming training in VR scenarios comprising dream-like elements, classical lucid dreaming training or no training. We found that VR-assisted training led to significantly stronger increases in lucid dreaming compared to the no-training condition. Eye signal-verified lucid dreams during polysomnography supported behavioural results. We discuss the potential mechanisms underlying these findings, in particular the role of synthetic dream-like experiences, incorporation of VR content in dream imagery serving as memory cues, and extended dissociative effects of VR session on subsequent experiences that might amplify lucid dreaming training during wakefulness. This article is part of the theme issue ‘Offline perception: voluntary and spontaneous perceptual experiences without matching external stimulation'.


Questionnaires
The following tests and questionnaires were used within the study: During screening, participants were asked for demographic data and their general lucid dreaming frequency in the form of the respective question of the MADRE. During the baseline week, participants filled out the MADRE to test for self-assessed (lucid) dream frequency of the past; PSQI for sleep quality; rMEQ for chronotype; FMI for trait mindfulness; LOC for trait locus of control; and PRMQ for trait prospective memory (self-or externally-cued). For research unrelated to our main study questions, further the PANAS, BDI, AUT, RAT, VIT, SPT, IPT were administered.
Over the 6 weeks of the study, participants day filled in a daily questionnaire comprising the VAMS (evening and morning), a brief dream report and the DLQ (morning); and for the VR condition and additional question "My dream was related to a virtual reality environment I have experienced" with a 6-point Likert scale (morning). The VAMS was collected for a different study, and thus not analysed.
During the final week, participants finished the same questionnaires/tasks as in the first week, however without demographic data, PSQI, rMEQ, PRMQ.
During a 4-week follow-up, participants filled out the PSQI to evaluate any effects of training on sleep quality; further the MADRE, FMI, LOC, PANAS, BDI. The PANAS and BDI were collected for a different study and thus not analysed.

Lucid dreaming training: explorative analyses
Consistent with the DLQ results, we did not observe a significant relation between the average VR dream incorporation over the 4 training weeks increases in the LuCiD insight scores, r = 0.33, p = .138. Given major inconsistencies in MADRE scores of several participants due to technical issues or poor compliance, we could not test for differences in this rather coarse measure of lucid dream frequency over more extended periods.
We performed several explorative tests to elucidate potential factors influencing the training effect. Testing if trait mindfulness (potentially influencing the attention payed to the current state of mind) or self-or externally-cued prospective memory (potentially influencing the intention to recognize upcoming dreams as such), we did not observe any significant correlation between these parameters and lucid dreaming increases, either for both training groups separately or combined (see Supplemental Figure S3).
To test if training increased particularly sporadic full-blown lucid dreams rather than a general level of lucidity in every night, we counted the amount of dream reports whose DLQ score exceeded 1, 2, 3, or 4 standard deviations beyond the individual mean of the baseline week. We found significant differences only for DLQ scores exceeding 4 SD of the baseline: F(2,36) = 5.481, p = .008, η 2 G = .233; see supplemental figure S4).

Reality Checks
To test if the training groups quantitatively differed in the intensity of their reality testing, in particular if the additional VR sessions increased the number of reality checks, we compared the daily logged reality check counts between training groups. We did not observe any differences between groups on the averaged total amount of checks, with VR participants performing 8.00±1.2 and active control participants 7.95±1.65 reality checks pay day; t(20) = -0.0813, p = 0.936.

Questionnaires
For questionnaire data at different time points not presented in the main manuscript, see Supplemental Table S1.  Figure S1: Association between baseline lucid dreaming as measured by the DLQ and training increases (see main manuscript).
Supplemental Figure S2: Association between dream incorporation of VR elements and lucid dreaming increases as measure by the DLQ or LuCiD (see main manuscript).
Supplemental Figure S3 (next page): Mindfulness, internal locus of control, prospective memory as potential predictors of training success. Neither the mindfulness (as measured by the FMI), nor internal locus of control (as measure by the LOC), nor self-(PSS) or environmentally-cued (PSE) short-term prospective memory significantly predicted lucid dreaming training gains as measured by the DLQ. Figure S4: Analysis of training effects on full-blown lucid dreams as measured by extreme DLQ scores. Tested are counts of lucid dream reports whose DLQ score exceeds 1, 2, 3, or 4 SD beyond the individual baseline (i.e. the first week). The most extreme DLQ values show the strongest training effect.

Lucid dreaming training
Participants of both the VR training group and the active control group received the same instructions for daily lucid dreaming training. In an individual introduction session, they were introduced into the concept of critically questioning their current state of mind to increase awareness of the discrepancies between real-life events and the strange events of dreams, thus serving as a trigger for achieving lucidity.
Specifically, they were instructed to ask themselves 5-10 times a day "Am I dreaming or not?", particularly in situation that subjectively felt unusual, dream-like, or strongly emotional. Participants of the VR training group were asked to do this in particular during VR sessions. All participants were further instructed not to answer this question mindlessly, but rather try to convince themselves that they are in a dream, and to look around for any strange things or inconsistencies that might be indicative of a dream. In addition, they were asked to perform "reality checks" to reliably test their current state of mind, for example trying to remember what happened just a few seconds ago. They received a "probe card" with the printed question "Is this a dream?", which they were instructed to observe closely for any inconsistencies in spelling or style; and look on the backside and frontside again to check if anything changes during the process. They were further instructed to choose specific times/occasions such as "while sitting at breakfast" or "while riding the bus to work" serving as regular cues for critical questions, and to log their reality checks using their smartphone after performing them. Throughout the experimental periods, participants received daily reminders via email, and personal reminders when they missed logging.
In addition, participants were instructed to go to sleep with the intention that upcoming dreams will be lucid; to carry out a particular action while dreaming (e.g. try to fly); and to move their eyes leftright-left-right when realizing to be in a lucid dream, try to move your eyes left-right-left-right.
Finally, participants of the training group were provided with written information and instructions on the subject of lucid dreaming, taken from the book "Exploring the world of lucid dreaming" (LaBerge & Rheingold, 1991). This specifically included the sub-sections 'Introduction to Lucid Dreaming, Dream-signs: Doors to Lucidity, The Dream-sign Inventory, Lucid Dreaming Induction Training' in addition to an additional section titled 'Instructions for you' written explicitly for the participants.

VR scenarios
Our starting and mainly used VR scenario was the custom made Spinoza Café, which was developed on the basis of an earlier version of a virtual replica of the university cafeteria (Ritter et al., 2012), hence a recognizable place for our student participants. Participants were tasked with serving customers their food and drinks, and cleaning up afterwards. Although the cafeteria itself is quite large, we opted to section off a space with only 4 tables, so our participants can reach everything within walking distance of the 5*5m 2 VR lab space. The scenario allowed researchers to manually or automatically spawn in customers. Once seated, participants used Vive controllers to pick up customers' orders at the counter and deliver it to the right table. Once the customers are finished, they left behind an empty cup and/or plate that had to be cleaned up before a new customer could sit in that spot. Empty cups had to be thrown in a bin, plates and trays had to be put on a little cart. Meanwhile, researchers could trigger different dream-like events, with an increasing level of surreality: change posters on the wall; change the clock on the wall to either fast forward or rewind; shuffle table numbers; move the trash bin (to different spots, but also inverted on the ceiling); male all customers stare at the participant; change all customers into mannequins; turn off gravity. All events happened behind the back of the participant to prevent jarring pop-in effects and increase the dream-like quality of the experience.
To keep the training engaging and ensure that participants had highly variable experiences just as during actuals dreaming, the following VR scenarios were used in addition.

Report 1
Report sent via official project email address on the following morning. Verbatim: ("Hey, I will bring the EEG at 11:30, I actually had a lucid dream! I hope I managed to move my eyes right but I think I did.").       [1 0 0 2 1 2 2 0 1 2 1 1]

Supplemental discussion
A confounding factor that should be noted is an apparent bias resulting from 'enthusiasm' drift, which might have masked actually stronger increases in the two training groups. Whereas we would have expected an increase in lucid dreaming even in the passive control group just by paying attention to this phenomenon on a daily basis, DLQ scores declined in 11 of 13 participants of the PC group. The most plausible explanation we could arrive at to explain these decreases was the presence of remarkable enthusiasm and optimism relating to the aims of the study (becoming more lucid) which inflated lucidity scores in the initial weeks; perhaps through an understandable combination of positive and wishful thinking on the part of study participants. The tapering-off in DLQ scores could have therefore been a gradual regression towards an 'accurately' reportable baseline, as participants became fatigued, disenchanted and generally despondent that the training would produce results. Needless to say, this implies that VR and AC groups may have been similarly effected; with measures training increases consisting of those which stood out above such refractory deflation. It could alternatively have been the case that the successful VR and AC participants had their biases further fuelled; with self-exaggeration of scores being procedurally encouraged as training milestones were reached.
These considerations were explored and tested at the end phase of this project; supporting many of our intuitions about the data collected. A number of participants were re-contacted some weeks after concluding their involvement, and had their own DLQ dream diary entries stripped of identifying temporal and other information, randomised in order, and sent back; with instructions to re-fill the quantitative numerical components based on their written and personal recollection of the dream. This produced strong correlations with the original reports (Cronbach's Alpha = 0.43, 0.465, 0.633, 0.948; p = 0.04, 0.02, 0.002, <0.0001 respectively) justifying-at least in principle-the mnemonic validity and accuracy of retrospective dream lucidity reappraisal. Most importantly, this provided a relatively bias-free re-examination of participant' impressions of their lucidity, controlling for variation in mood, enthusiasm and optimism over the 42 days. One VR participant whose data reported a severe negative trend, upon re-analysis, saw remarkable gains (see Supplemental Figure  S4); an observation backed up by the participant's own anecdotal reports of having "significantly more" lucid dreams as result of the intervention. Additionally, the PC participant whose data showed the most severe decline, upon re-evaluation, produced a near-uniform flat gradient-as one would expect from a passive control condition. Without comprehensive re-evaluation on a larger sample size, further conclusions are difficult to draw; however it does appear that longitudinal data collected on a highly subjective, fundamentally oscillatory/noisy quanta such as dream lucidity is apt to be overtaken by larger shifts in participants affect, at least in certain individuals. Future studies would do well to control for this possibility.
Supplemental Figure S4: Validity comparison between originally and postspectively revisited (order randomised, completed on a single sitting) DLQ questionnaires. Despite showing opposite trend-lines, both sets correlate considerably (Cronbach's Alpha = 0.465, p = 0.018). When the bias was corrected for through subtraction of linear estimations, both sets correlated significantly more (Cronbach's Alpha = 0.834, p < 0.001) indicating highly accurate lucidity estimations from postspective dream diary analysis; compensating for longitudinal affective drift. In this case from the VR training group, a potentially positive trend over trainingconsistent with the retrospective appraisal by the participantmight have been masked by some sort of drift, resulting in a seemingly negative trend over the course of the study.

Further supplemental data
Further supplemental data and analysis scripts can be found on https://osf.io/jrph2