The Effect of Orientation on the Readability and Comfort of 3D-Printed Braille

Fused Deposition Modeling (FDM) is a low-cost method of 3D printing that involves stacking horizontal layers of plastic. FDM is used to produce tactile graphics and interfaces for people with visual impairments. Unfortunately, the print orientation can alter the structure and quality of braille and text. The difference between printing braille vertically and horizontally has been documented. However, we found no comprehensive study of these angles or the angles in between, nor any study providing a quantitative and qualitative user evaluation. We conducted two mixed-methods studies to evaluate the performance of braille printed at different angles. We measured reading time and subjective preference and performed a thematic analysis of participants’ responses. Our participants were faster using and preferred 75° and vertical braille over horizontal braille. These results provide makers with guidelines for creating models with readable 3D-printed braille.

The angle at which an object is 3D printed impacts its geometry and surface quality, as demonstrated by the illustration on the left.This quality diference may infuence the tactile properties of braille, afecting its comfort and readability.Therefore, we conducted two experiments evaluating the efect of printing braille at diferent angles.For one experiment we sanded the prints and for the other one we did not.We taped braille prints onto a stock paper page and recorded participants' responses, including reading time and preference, as seen in the illustration on the right.

ABSTRACT
Fused Deposition Modeling (FDM) is a low-cost method of 3D printing that involves stacking horizontal layers of plastic.FDM is used to produce tactile graphics and interfaces for people with visual impairments.Unfortunately, the print orientation can alter the structure and quality of braille and text.The diference between printing braille vertically and horizontally has been documented.However, we found no comprehensive study of these angles or the angles in between, nor any study providing a quantitative and qualitative user evaluation.We conducted two mixed-methods studies to evaluate the performance of braille printed at diferent angles.We measured reading time and subjective preference and performed a

INTRODUCTION
Three-dimensional (3D) printing has been explored as a relatively inexpensive and customizable medium to produce tactile interfaces for people with visual impairments.Examples range from tools in education [6,7,38], orientation and mobility [19,24], and medicine [1,14].Relevant textual or auditory information must often accompany these interfaces to make them usable [45].There are many advances in technologies to label such models like infrared tags [13], Quick Response (QR) codes [2], and clicking devices [48] that all trigger audio labels.However, while using braille has limitations, its familiarity and universality make it an accessible system for presenting contextual information.Therefore, it is important to understand the benefts, limitations, and best practices of 3D printing as a medium to produce braille.While there are many types of 3D printing, Fused Deposition Modeling (FDM) is considered among the most widespread and economically accessible [12].FDM 3D printing works by stacking horizontal layers of molten material, most often plastic, meaning that a model is created from horizontal slices.Therefore, a print's structure and surface quality are related to print orientation [8].These challenges can impact the readability and comfort of braille.Fabrication techniques, which include 3D printing, have had a meaningful impact on the development of general assistive technology, including tools for people with visual impairments [17,18].For example, Hurst and Tobias explore the potential of Do-It-Yourself (DIY) techniques via case studies [26].They describe modern 3D printers as comparatively inexpensive tools, and more accessible than other traditional fabrication methods.They also praise their cultural impact and community access in public spaces like libraries and disability centers.Similarly, Buehler et al. investigated how online communities share these models, including tactile graphics and other tools designed for people with visual impairments in mind [5].The space to share models and ideas is considered one of the strengths of 3D printing.
A white paper by the DIAGRAM Center noted the diference between vertical and horizontal braille based on the printing angle [10].In both cases, the braille is added onto a back surface akin to a sign, referred to as a plate in our experiment.For vertical braille, this plate is printed to stand vertically upwards with braille to its side.Correspondingly, a plate with horizontal braille is printed to lay fat with the braille cells at the top.Figure 2 illustrates the model of a braille dot printed vertically on the left and horizontally on the right.They describe how vertical braille was preferred over horizontal braille.However, the report provides only general recommendations and no details about the results of an empirical evaluation.In this work, we expand upon the DIAGRAM Center's fndings to investigate the relationship between print orientation and reading experience with a more granular evaluation and analysis.Specifcally, we conducted two mixed methods experiments that measured participants' preference and performance for stimuli 3D printed at diferent angles.
These experiments were designed to answer the following research questions.
• RQ1 What is the impact of printing angle on the reading speed of 3D-printed braille?
Figure 2: The print on the left is oriented at 90 degrees from the plate, whereas the right one is at 0 degrees.The former has to deal with overhangs for the frst several layers, which are printed partially above the air instead of being supported by the lower layer.
• RQ2 How do printing angles afect readers' reported ease of discerning individual characters?• RQ3 How do printing angles afect readers' reported comfort when reading the 3D-printed braille?We evaluate and report quantitative metrics, namely reading time and Likert preferences, and report a qualitative analysis of common themes brought up by participants during the experiments.The two experiments test sanded and not sanded, referred to as unsanded, stimuli.Namely, sanding allows us to ask participants to read fast while minimizing the risk of injury, whereas not sanding provides a "straight-from-the-machine" perspective.The purpose of the study is not to compare the impact of sanding across both experiments but to provide a multifaceted analysis by exploiting the strengths of each design.
To summarize, this paper contributes: (1) the design and results of two studies comparing the usability of 3D-printed braille at various angles: (a) a controlled experiment in which we measured the reading speed at which participants read braille and (b) semi-structured interviews and Likert-style questionnaires in which participants rated reading character discernment and comfort.(2) design and implementation recommendations for using braille in 3D-printed interfaces based on these results.

BACKGROUND AND RELATED WORK
We frst provide a summary of FDM 3D printing, which is the specifc technology we are investigating.Then, we recapitulate the relevance of its applications in tactile graphics and other interfaces.Finally, we explore similar evaluations of 3D-printed braille.

FDM 3D Printing
Fused Deposition Modeling is a method of 3D printing that consists of creating a model by building upon it in horizontal layers [42].A nozzle is attached to an arm that moves in continuous paths where it deposits the molten material on its trail.Note that higher layers in the model rest on lower ones to prevent warps or defects from overhangs.Many experts such as Prusa Research, our printer's manufacturer, suggest avoiding overhangs angled at less than 45° from the horizontal [47].45° is geometrically relevant since it is the angle at which at least half of the upper layer makes contact with the lower layer, meaning better adhesion and less impact from gravity.
Weeren et al. identifed a few problems that afect the surface quality of FDM printed objects [54].They describe how the horizontal slicing creates a staircase efect from layers stacked upon each other.This staircase efect is diferent based on the slope of an object [52], as illustrated in fg. 2. This means that its orientation during print will impact the dimensional accuracy-how close the object resembles the theoretical model-and the surface roughness.Weeren et al. also describe how defects at the start and end of the nozzle's path lead to imperfections on the surface.For example, the gaps in between braille dots might have small nubs from plastic that increase the roughness of the text.Among other physical limitations, the size of the nozzle constrains the width of a line of extruded material [23,47].FDM printers have lower resolutions and surface quality than other printing methods [27].However, FDM is still considered one of the most afordable commercial types of 3D printing, including initial equipment purchases and per-model costs [12].

3D-printed Tactile Graphics and 3D Models
Tactile graphics are tactile representations of two-dimensional images, such as textbook graphics in STEM education and mathematics testing [30].However, they can also have recreational applications like in art [32,57].There are many documented guidelines and good practices for developing tactile graphics [15,44].3D models often aim to represent something inherently embedded in 3D space, like monuments, buildings, or chemical particles [53].Holloway et al. evaluated some diferences between tactile graphics and 3D models for orientation and mobility maps [25].
Over the last decade, several studies have focused on understanding the educational uses of 3D printing for accessible graphics and models [28].Buehler et al. identifed some of the benefts and limitations of 3D printing in special education classrooms, including for children with visual impairments [6,7].Similarly, 3D printing methods have been used in other contexts such as for the production of orientation and mobility tools [24], the customization of circuit engineering instruments [11], and labeling medicine [1,14].3D printing also provides customization that allows to design for diferent visual conditions [42].For example, Gotzelmann proposes a combination of a smart device and their 3D-printed graphic to visually enhance parts of tactile graphics for users with low vision [19].There is also work addressing the limitations of Computer-Aided Design (CAD) modeling software, which is often used in 3D printing [40,49].These examples highlight the use and potential of 3D printing as a medium to create tactile interfaces.

Labeling
Methods.The use of braille in tactile graphics is standardized, and organizations like the Braille Authority of North America (BANA) have guidelines for it [44].Among others, the Round Table advise on the benefts and limitations of diferent tagging techniques including braille, basements (which are separate signs containing labels and a model outline), and audio notes triggered by QR codes or NFC tags.Similarly, the 3D Printing for Visually Impaired (3D4VIP) project coordinated by the Royal Dutch Visio recompiled standards and best practices for the use of 3D printing for tactile graphics [55].Namely they specify important parameters to consider when 3D printing models including the importance of print orientation.However, they also do not provide an empirical evaluation or mention angles beyond horizontal and vertical braille.
There are some challenges to the use of braille in tactile graphics and 3D models.For example, braille cannot be resized (which prevents re-scaling models), it takes a signifcant amount of space, and it requires that users know how to read it.These can impose constraints on the text length that can be added onto models.However, the community has developed several tools and methods to circumvent some of these limitations.For example, devices like The Tactile Talking Tablet allow users to mount raised graphics onto a tablet for auditory feedback [29,30].Baker et al. proposed embedding QR codes to encode audio labels that could be read with another device, such as a smartphone [2].However, they also found that braille-literate participants preferred braille and read it faster.Shi et al. propose Tickers and Talkers [48], which use physical triggers to encode audio labels for 3D models which may not have large fat areas to include braille or QR codes.Other methods, such as The Tactile Graphic Helper by Fusco and Morash [16], use computer vision to identify where a user might be pointing.While these methods can be very efective, many of them require an external device to use.In the case of devices that need to be connected to a computer, this can hinder its portability.Similarly, other programs may pose other software requirements that might not be viable in the long term or may need to be maintained.This also poses a barrier to entry via the use of cameras or smartphones, which, while expansive might not be available in some settings or adopted by all populations.Therefore, while braille has its limitations, there are also valuable reasons for its continued use.

Evaluations of Braille Displays
While outside the scope of this project, we note that many kinds of tactile displays are used for graphics and other texts [9,50,56].Morash et al. conducted an evaluation proposing one such interface.To evaluate the performance of their high density pin display, they tested the reading speed and reported the difculty of their various pin confgurations to emulate braille patterns with multiple pins.While their proposed display was not 3D printed, the stimuli they used for their studies were.Minatani explored the usability of 3Dprinted braille for embossing tactile graphics, namely geographical maps [39].They proposed a design pipeline and highlighted some of the practical obstructions they faced in their prototyping.Namely, they describe problems with the inferior surface quality and the readability of city labels.Loconsole et al. compare two low-cost FDM 3D printing techniques for producing braille as well as with a professional printer [33].The techniques consist of the layer-bylayer FDM approach discussed above, and a novel method named continuous fow.For this method, they programmed an FDM printer to continuously extrude plastic by moving up and down in between layers to avoid residual flament in between cells.A study of their method seemed to improve the comfort of users reading braille, but they highlight how commercial software often limits the ability to program arbitrary pathways.Neither of these evaluations consider print orientation and the ones using FDM printers describe the high surface roughness of the braille.
Finally, the DIAGRAM Center conducted a series of studies to determine standards for adding braille to tactile graphics, including testing printing [10].They tested a variety of printers to determine that braille printed perpendicular to the printing bed, which they call vertical braille, was better than braille printed parallel to the printing bed.In their report, they mention testing other angles and recommend avoiding them, but no further explanation is devoted to this argument.

METHODOLOGY
Our study consisted of two within-subject experiments with similar stimuli and study conditions.Both experiments were comprised of a subjective Likert scale and qualitative evaluation of the impact of angle on participant responses.The frst experiment contained sanded stimuli and also used reading speed as a proxy for performance.In the second experiment we did not sand the stimuli, and so we refer to it as the unsanded evaluation.This separation was made as initial pilot studies suggested that some of the prints straight from the printers would have nubs that could hurt participants when asked to read as fast as possible.
We tested seven print angles from 0 degrees (horizontal) to 90 degrees (vertical) from the print plate in 15-degree intervals and included a control consisting of braille embossed on 80# stock paper and produced by the National Braille Press in Boston, MA.For both experiments, participants did not know the angle of print or sentence of a presented plate.All the plates were printed with removable supports so that they could be laid down fat after printing.A fexible casing was also made to hide the slope in the plates printed at angles, and this casing was glued onto a piece of stock paper so that participants could easily fnd and move the stimuli.The supplemental materials of this study can be found in https://osf.io/t2rbq/and the preregistration in https://osf.io/mcyv5.

Participants
We recruited thirteen adults through the Carroll Center for the Blind.One was excluded from the analysis since a stimulus was presented to them more than once.Of the twelve participants, seven self-identifed as female and fve as male.Participant ages ranged from 29 to 81 (median= 59).Ten participants identifed as having complete or total blindness; one identifed as "almost completely blind with some light perception", and another one as having signifcantly low vision.Similarly, eleven participants started reading braille before the age of eleven with the other at 38. Experimental procedures were approved by our Internal Review Board, and informed consent was acquired from all participants.Participants were compensated with a $40 US gift card.

Reading Stimuli
The braille was modeled in OpenScad by adjusting the specifcation of existing code found online [51].For the dimensions, we followed BANA's guidelines for signage which are shown in table 1.We used the smallest values in each of their intervals as pilot studies suggested that higher dots could be more uncomfortable.The braille dots were modeled as hemispheres with truncated tops to make  1: Dimensions of braille for signage used to create the stimuli.From [43].
them fat.The characters were added on top of a rectangular plate with a slope on its long side for the angles to print.The models were then exported as STL fles and added onto a slicer software (Prusa Slicer) to generate the instructions for how to print them.To avoid the plates warping in our experimental stimuli, we add support structures. Figure 3 exemplifes the stimuli used and shows the printing angle.
3.2.1 Technology.Three Original Prusa Mk3S+ printers were used to print all the stimuli from a Polylactic Acid (PLA) flament.Given the randomization we used, all 64 angle-sentence combinations were printed twice, once for the sanded and another time for the unsanded experiment.The specifc print parameters and rationales are included in the supplemental materials, which can be found at https://osf.io/t2rbq/.

Sentence Selection.
To minimize the efect of context among diferent stimuli, we chose a set of standardized sentences used in braille reading speed assessments [31].We used an extension of the sentence corpus for the MNREAD acuity charts [35].While these sentences were originally proposed to measure visual acuity [36], they are often used in the evaluation of braille reading performance [41].The extension consists of computer-generated sentences from 13 diferent templates [34].They all contain the same number of characters, simple vocabulary, and no punctuation.To maintain sentence diversity and minimize the impact of context, we selected sentences made from diferent templates.
MNREAD sentences also provide a standardized spacial layout, which we slightly adjusted to better ft the dimensions of our printing technology.Instead of having the text in three lines, we found a subset of the sentences that could ft in two lines without cropping words, which allowed us to print four to six plates per printing session.Legge et al. suggested that characters per second was an appropriate metric to measure braille reading speed for these sets of sentences [31].Hence, to maintain a uniform number of characters and a consistent spatial layout, we decided to use grade 1 (uncontracted) braille, similar to the evaluation by Morash et al. [41].  .We note that while some visual diferences can be seen in the texture and shape of the dots, the most important diferences are tactile.

Randomization.
Despite the MNREAD-like sentences providing a standard for uniform reading speed, we wanted to minimize the impact of specifc sentences on the evaluation of the angle.Therefore, we randomized sentence and angle combinations so that a unique pair only appears again once every other combination has already been used.Since we had eight conditions including the control, we chose eight sentences to pair them.That means that the frst group of eight participants did not share any sentence angle combination between themselves, and neither would the second group among themselves.Finally, we also randomized the order in which these were presented to minimize ordering efects and the impact of reading fatigue.Note that we do not randomize which plates are sanded, since the goal is not to measure the efect of sanding.The specifcs and randomization code are provided in the preregistration and supplemental materials found in https://osf.io/t2rbq/.
3.2.4Sanding.For the frst experiment, we evaluated the plates after being sanded and we measured the reading speed of participants.Sanding is considered one of the best ways to improve the surface quality of 3D-printed objects [58].This consisted of passing a 320 grit sandpaper over the stimuli back and forth fve times.

Procedures
3.3.1 Sanded Evaluation.For the frst experiment, participants were prompted with the same tutorial plate to get them adjusted to the medium and the uncontracted braille.The angle for this plate was 45° as it was the median angle in our set.After briefng the experiment and receiving consent, we recorded the audio of the participants reading out loud.We presented the plates one by one and asked participants to withhold from reading once they had found the stimuli.Then we instructed them to begin reading after a countdown.In between trials, participants were asked to provide a rating on the discernment and comfort of these plates on a Likert scale from one to seven.
We also asked open-ended questions for qualitative feedback.This served to extend the time between trials to minimize reading fatigue.We looked at the recordings' audio waveforms to measure reading time as the diference in seconds between the completion and instruction timestamps.

Unsanded
Evaluation.This experiment was conducted immediately after the sanded evaluation.Participants were told to read the sentences as much as they were comfortable.We presented the stimuli like in the sanded experiment.The Likert scale consisted of two questions with seven response options each (1-strongly disagree, 2-disagree, 3-somewhat disagree, 4-neutral, 5-somewhat agree, 6-agree, 7-strongly agree).The Likert questions we used were the following: • It was easy to discern the individual braille characters.
• The braille characters were comfortable.

Trial count.
Each participant saw a total of 16 plates, two for each angle, once for the sanded experiment and once for the unsanded.Hence there were 192 total trials across all participants and the two experiments.They also saw a tutorial plate for which data was not recorded.

Analysis
We conducted non-parametric repeated measures analyses for the reading time and the Likert data since the normality and sphericity assumptions were unmet.We used Friedman's test to determine whether the angle signifcantly impacted the metrics we evaluated.
If so, we ran pairwise Wilcoxon Signed Rank tests.Specifcally, we ran one tailed tests to determine if the higher angles performed better than the lower angles.We adjusted -values with the Benjamini-Hochberg False Discovery Rate [3] and used a 0.05 signifcance level.Finally, we used the Pratt method [46] to account for ties in the Wilcoxon tests.All analysis code was preregistered before conducting the experiment, except for the Pratt tie-breaking method as we did not anticipate ties in our results.The preregistered analysis code can be viewed on OSF at https://osf.io/mcyv5.
For the qualitative analyses, we transcribed, compiled, and cleaned the audio recordings of each session.We then encoded responses into general themes using an inductive thematic analysis approach [4].Finally, we also separated participant's comments about plates for specifc angles to further highlight the impact of angles on results.

Hypotheses.
• (H1) Participants are faster at reading braille printed at higher angles, except for braille printed horizontally.
As per the study conducted at the DIAGRAM center, we expect vertical braille, to perform better than horizontal braille [10].• H2Participants rate braille to be more comfortable at higher angles.We also expect higher angles to perform better.• H3 Participants rate braille to be more discernable at lower angles.While comfort is a factor of readability, we foresaw a possible inverse relationship between comfort and discernment.For example, Prusa Research suggested printing half spheres horizontally for greater dimensional accuracy, as this minimizes overhangs.• H4 Participants are faster at reading the control print over the 3D-printed plates.Other explorations of 3Dprinted braille determined that FDM printing can produce lower quality braille than achievable with traditional embossers or high-grade professional printers [33,39].

RESULTS
In the following section, we report the reading time, comfort, and discernment results for both experiments as outlined by our hypotheses.
4.2.2Unsanded Evaluation.Fig. 6 shows the distribution of Likert ratings for the unsanded experiment separated by angle for comfort and discernment.For the comfort ratings, we found that for each pair of angles, the higher one was most often signifcantly better than the lower one.We note a few exceptions, 15° (median = 1.5) and 30° (median = 2) were not found to be signifcantly better than 0° (median=2).On the other hand 90° (median= 7) was not found to be better than 75° (median = 7, = 0.56).Note that 4 is the neutral response in the Likert questionnaire.For the comfort ratings, the medians of all angles under 45° are lower than the neutral response.The evaluation of discernment ratings shows similar trends.For discernment, the control (median = 7), the 60° (median = 7), 75° (median = 7), and 90° (median = 7) plates performed signifcantly better than the 45° and lower counterparts.We also highlight how 45° (median =4.5) is rated more discernible than 30° (median =3.5, = 0.014).

H4: Control
We used the same tests to determine if the reading time for the control (median = 11.75) was lower than the other angles as proposed in H4.We found no statistical signifcance that this was the case for any of the angles.Conversely, the control was determined to be signifcantly rated as more comfortable than 0° ( = 0.013), 15° Figure 5: The median Likert scores, per angle, for comfort on the left and discernment on the right for sanded plates.Note that a one on the x-axis means the plate was deemed very uncomfortable, and a seven means that the plate was considered very comfortable.Finally, the dashed line at four on the x-axis represents the neutral response (neither comfortable nor uncomfortable).The thick error bars represent 68% bootstrap confdence intervals while the thin ones represent 95%.( = 0.013), 30° ( = 0.014), 45° ( = 0.014), and 60° ( = 0.0163) in the sanded experiment.The unsanded experiment mirrored this pattern with the control being rated signifcantly more comfortable than every other angle including vertical braille ( = 0.037) but not 75° braille ( = 0.061) although the later was close to the threshold.

DISCUSSION
In the following section we discuss how our experimental analysis answered the research questions presented in section 1.

RQ1: Impact of Angle on Reading Speed
We found a signifcant impact of angle on reading speed.We hypothesized that this relationship would be strictly increasing; however, the pattern seemed to revolve around two groups of diferent angles.We found that angles printed above 45° performed better than their lower counterparts, with little diferences within them.The reading time analysis did not detect signifcant diferences between 75° and 90° .These quantitative results were supported by participant responses regarding how they interacted with the diferent plates.Participants described the importance of their fnger movements, namely "gliding", and how some plates impeded that in various ways.P1 mentioned their "fngers were just snagging" on the 15° plate.Other participants, like P11, discussed the importance of "not catching on my [their] fngers in the rough corners".P11 described the 30° plates as having "a little drag coming back".Similarly, P0 mentioned this efect only when going backward for the 15° plate.They said: 'I noticed that I'm picking up my fngers ."P6 shared this sentiment by expressing how "if I go this way, it's rougher, but if I'm reading from left to right, then it's not the same" for the 30° plate.Similarly, participants also talked positively about how some plates were more consistent in both directions.P10 described the dimensions of the 60° plate as: "dot height was fne, spacing excellent, braille much easier to move both directions." We also note the diferences in the variance of the data across angles as seen in the confdence intervals of fg. 4.This variability is common in diferent people's reading speeds [37].However, we call attention to how the group of 60°, 75°, and 90° printed plates were more consistent at afording faster braille reading.
In section 2.1 on FDM 3D Printing, we mentioned the signifcance of 45° from a printing perspective.Namely, we explained how angles lower than 45° may lead to overhangs that afect printing quality.Similarly, the staircase efect for these angles is more pronounced, as the printer takes longer steps between layers for low angles.We propose these are reasons some of the braille afect participants' motions across the braille.
5.1.1Performance of the control.H4 tested the relative performance of participants between the 3D-printed braille and a paperprinted control.Our reading time analysis indicated that participants were not faster at reading paper braille than the 3D-printed one.However, the qualitative and Likert scale evaluations indicated that participants had a notable preference for the paper braille.Therefore, we hypothesize that since the braille types were diferent, participants might have lost time adjusting to the diferent dimensions and textures of the paper braille.

RQ2 and RQ3: Impact of Angle on Participant Responses
In accordance with the results from the DIAGRAM center, we found vertical braille to be preferred over horizontal braille [10].Participant responses for both comfort and discernment mirror the trends seen in the reading speed evaluation.

Comfort.
For comfort, we generally found increasingly higher angles to perform better than lower ones across both experiments.However, we also found there to be a jump between the angles lower than 45° and those higher.The sanded and unsanded experiments show similar trends, with them being more pronounced in the unsanded experiment.We do highlight that across both experiments the 75° plate is rated to be as good as vertical braille.This is supported by participant responses that characterize lower angles as "rougher" and "sharp", and higher angles as generally smoother.
Comparatively the lack of roughness was considered a positive, P12 described the 75° unsanded plate as: "not rough, looks gorgeous".While it was less common, a couple of participants also noted on the texture of the plate itself.P5 described the backdrop of a 75° plate as "shiny" and P6 described the backdrop of the the 15° plate like "curdoroy".
In relation to smoothness, P0 characterized the importance of matching the expectations of users as demonstrated by them saying "If you're expecting smooth Braille, but you get like hard braille, that sure is not ideal".On the other hand, P7 mentioned being familiar with rough displays: "I'd say it's about the consistency of a hard braille display".This sentiment even applied with the study itself, with how P11 described the unsanded version of 75° as "smooth, I was expecting to get, you know, Mr. Scratchy right of the bat. . .this is nice." We propose that plastic defects and deposits in the layer's start and end points are major contributors to the roughness of this braille.For horizontal braille, the starting point of a braille cell will always be to the side of the cell, which is where participants run their fngers.On vertical braille, the starting and ending points of a braille cell are always against the plate, which is more hidden from the fngers.This could explain why participants felt more roughness on some sides than on others, as they could have been bumping on starting points.

Discernment.
We hypothesized that there could be an inverse relationship between discernment and comfort.Some comments from the participants alluded to this idea, namely with some lower angles described as "proud" and having "good height".Similarly, some made comments about the texture inconsistency of some higher angles, alluding to them having less dimensional accuracy which aligns with the expectations of layered printing.For the 90° plate, a P1 mentioned "So I like this one a lot.It's smooth and comfortable to read, but I do feel like it was a little bit difcult to discern, because some of the dots looked a little bit too sanded, maybe like not raised enough".Nevertheless, it generally seemed that these effects were not as noticeable to most participants.Conversely, other factors like surface roughness were a bigger determinant of reading time.
We found that a few participants described many of the plates as being rough but not unclear.For example, P0 described a plate as having "some roughness to it, but the letters are all clear." A few participants also described instances of what they called ghost dots as mentioned by P5.They described this a a dots that were hard to tell if they were there.Another participant described it as "almost missing the dot three . . .it's almost not there''.A similar sentiment was described by P6 who mentioned that the dots seemed to recede for the tutorial.Finally, a couple participants described braille dots as "proud".P11 defned this term as "the word we tend to use for like the height of the dots."Similarly, P1 described the braille as "crispy" and defned it as "when we say crispy braille, it's just a, you know, really defned and really clear and you can read it really well".However, dot height was also described negatively.P12 mentioned that the feel of the 45° plate was "horrible horrible...The dots are taller but they're they're rough." While not the emphasis of the work, we note that the trends for both the sanded and unsanded experiments are similar, hinting at these structural diferences of the braille.The trends for both comfort and discernment across experiments are very consistent with each other, the main diference being the sanded plates having higher medians, which is a reasonable expectation.

Limitations and Future Work
Below we discuss some of the limitations, exploratory analysis, and future work for this project.

Reading Time.
First, we note how the method for measuring reading speed also accounted for reaction time and other cognitive processes.Other evaluations utilize more precise ways of measuring reading time, such as following fnger paths with trackers strapped to participant's hands [41].We believe the trade-of provided participants with more movement and comfort.However, other fnger-tracking systems might make the time measures more accurate.

Text length.
These results may only generalize to shorter texts due to reading fatigue.Some participants explicitly mentioned the concern for reading plastic braille for extended periods.For example, P5 described how they " wouldn't want to read a whole page of signage with this" when talking about the 0° plate.P9 echoed this sentiment when describing the 60° plate as "not uncomfortable, but it is not something I would read for a long time either".Conversely, P1 mentioned their preference for the 75° plate in the context of museum plaques.Specifcally, they mentioned, "If you put these in museums or something, sometimes it's usually a lot of information.I would not wanna read it for more than a couple sentences, but this one, I would feel like I would be able to read it for like a page or something".3D printing to produce general braille has many limitations, including storage space and long printing times.However, the results of these experiments should be applicable for the use of labels and short texts, which is where we believe their applications mainly lie.

Experiment Separation.
We believe that separating the experiment by sanded and unsanded let us gain valuable information from each of the experiments.This sentiment was reiterated by P7 when talking about the roughness of the 15° plate: "glad you don't time this, the only reason I can read it, and I'm gonna completely honest, that I can read it accurately is because I had the last one [referring to the corresponding sanded plate]."While we believe sanding the plates was adequate to remove some of the larger protrusions from the braille, we also recognize how it afects the stimuli.In an evaluation of surface quality for the adhesion of a spray, Hanton et al. found that sanding seemed to make large surface defects smaller, but also increased the number of imperfections of the surface [21].We consider performing exploratory analysis to determine the impact of sanding on surface quality.

Other
Considerations.While 3D printing is described as an afordable manufacturing method, this notion of low cost is relative.For example, Gupta et al. emphasize 3D printing's potential as low cost in India [20].However, Zuniga-Zabala and Guerra-Gomez mention how the prices to access this technology remain unfeasible for some populations, especially in countries with lower minimum wages [57].The Prusa M3KS evaluated by Chen et al. in their cost and printing time comparison of printers was priced at around one thousand USD, which may not be an accessible cost for individuals and many institutions [12].Cheaper printers exist for around 200-300 dollars but are less user-friendly and require efort to get goodquality prints.This was also brought up by P0, who mentioned the durability of braille but their perceived cost: "I don't think that smooshes, I don't think that gets damaged very much, but it's probably much more expensive to do." 5.3.5 Exploratory Analysis.In a study to determine what factors impact braille reading performance, Martinello et al. found age when learning braille to be a signifcant factor [37].We believe this is one of the many demographic factors that could afect people's experience with 3D-printed braille.Namely, we propose to do some exploratory analysis on demographic correlates to see if there are noticeable diferences in experiences.This could inform further studies to better target people's specifc needs or preferences.Reading angle also has an impact in braille reading [22], something which was also brought up by a couple of participants.

Recommendations
First, we reiterate that there are many guidelines about the production and use of braille for tactile graphics and 3D models [10,25,45].We found that 75° and 90° printed stimuli performed best in both the reading speed and subjective analyses for the 3D-printed plates.The performance of vertical, or 90°, braille is consistent with the recommendations of the DIAGRAM Center.Printing vertically has many design implications, as any other structures that rise from the surface will have overhangs.This can have negative efects on the printing quality of those areas of the model.Nevertheless, printing at 75° can help minimize some of those sloped overhangs which means that designers have more wiggle room to make better models without sacrifcing braille quality.
(1) Identify beforehand when adding braille into a model is reasonable.Models that require diferent scales or that do not have fat parts might beneft from diferent tagging systems.(2) Braille should be printed at angles above 45°, but ideally as close to 90° as possible.Namely, we recommend 75° as an option, especially if this minimizes overhangs from other parts of the model.(3) Consider post-processing like sanding if other restrictions do not allow printing braille labels at good angles.

CONCLUSION
In this paper, we present two mixed-method user studies to evaluate the impact of printing angle on the usability of FDM 3D-printed braille.We found that participants read 60°, 75°, and 90° printed braille more consistently faster than braille printed at lower angles.We also found 75° and 90° braille to be rated as more comfortable and discernable than the other angles.Participants' qualitative responses complemented this analysis and highlighted how the plates for these two angles were smoother.While paper braille was not found to be better than 3D-printed braille in the reading time evaluations, participants showed a clear preference for it in the subjective evaluations.Nonetheless, many praised the potential applications of 3D-printed signage, mentioning applications like museum plaques, elevators, and hiking trails.Some also suggested the potential durability of 3D-printed braille over paper.These fndings present actionable advice for makers who incorporate braille in their designs, as there are more angle options to print without sacrifcing user comfort.

Figure 1 :
Figure1: The angle at which an object is 3D printed impacts its geometry and surface quality, as demonstrated by the illustration on the left.This quality diference may infuence the tactile properties of braille, afecting its comfort and readability.Therefore, we conducted two experiments evaluating the efect of printing braille at diferent angles.For one experiment we sanded the prints and for the other one we did not.We taped braille prints onto a stock paper page and recorded participants' responses, including reading time and preference, as seen in the illustration on the right.

Figure 3 :
Figure3: On the left (a) is a side view of the printer showing the plates at six printed angles (missing horizontal braille).On the right (b), we show two stimuli printed at diferent angles, namely 0° and 75°.We note that while some visual diferences can be seen in the texture and shape of the dots, the most important diferences are tactile.

Figure 4 :
Figure 4: Results of the median reading times (x-axis) of participants grouped by angle (y-axis).The thick error bars represent 68% bootstrap confdence intervals for the median while the thin bars represent 95%.

Figure 6 :
Figure 6: The median Likert scores for comfort on the left and discernment on the right for unsanded plates.The thick error bars represent 68% bootstrap confdence intervals while the thin ones represent 95%.

Table 2 :
Statistical distribution summary of reading times for the Sanded experiment.

Table 3 :
Qualitative evaluation for the sanded Experiment.Noticeable themes separated by angle.

Table 4 :
Qualitative evaluation for the unsanded experiment.Noticeable themes separated by angle.