Applications
Editor: Mike Potel
Design-to-Fabricate Maker Hardware Requires Maker Software Ryan Schmidt Autodesk Research Matt Ratto University of Toronto
C
omputing’s evolution is driven largely by the development of novel hardware capabilities, which can be fully exploited only by new types of software interfaces. This was particularly evident in digital fabrication’s early days, when the development of CNC (computer numerical control) machines at MIT led to an interest in CAD tools, essentially to “create content” for this new hardware. Ivan Sutherland’s seminal Sketchpad system—and hence the first GUI—was one fruit of this endeavor. Recent years have seen a significant leap forward in rapid-prototyping hardware, particularly in 3D printing. At the high end, 3D printers have micro meter resolution and can combine flexible, rigid, and transparent materials. Consumer-level systems have output quality similar to machines that cost tens of thousands of dollars just a few years ago. And mailorder 3D-printing houses provide affordable access to a range of materials, from metals to ceramics. Now that 3D-printing hardware has become vastly more accessible, there’s a growing community of makers looking for software that lets them create content for their new devices. Of course, such tools already exist, in the form of professional CAD software. However, much like early digital-fabrication hardware, these tools’ cost and complexity are beyond the hobbyist maker’s reach. Here, we take a look at how researchers have tried to overcome this problem by creating user-friendly software that assists the design of 3D fabricated objects. In particular, we focus on the Autodesk meshmixer toolset.
CAD Tools for Makers Traditional CAD packages aim to support the pipelines and practices of professional engineering and 26
November/December 2013
industrial design. In the face of these rigid processes, innovation in easy-to-use CAD tools largely ground to a halt outside of academic research. However, the democratization we’re seeing in digital-fabrication hardware brings with it a much larger community interested in designing their own 3D objects. Even if they’re professionals by day, when they put on the maker hat, they have the freedom and desire to explore novel, unconventional tools. So, digital fabrication provides an ideal playground for researchers and software developers—the tool makers—to experiment with new approaches to 2D and 3D design. This is spurring the development of new types of design tools, particularly those that deeply integrate the knowledge that the end goal is a physical object. Although academic computer graphics researchers have always been interested in easy-to-use design tools, research is increasingly incorporating fabrication constraints into the design process. These design-to-fabricate systems enable the rapid creation of complex physically realizable designs that would otherwise be well beyond even moderately skilled users’ capabilities. A notable example is Nobuyuki Umetani and his colleagues’ guided furniture design tool (see Figure 1).1 Users can, for instance, explore any number of wildly complex bookshelf designs, with the system providing subtle hints to guide a design toward one that’s physically stable and manufacturable. Some research has taken the reverse approach, applying geometric analysis to augment users’ realworld fabrication abilities. In the Sculpting-byNumbers project, Alec Rivers and his colleagues helped novices create clay sculptures of given 3D models.2 Using a camera and projector, the system analyzed the sculpture and displayed guidance information directly on it.
Published by the IEEE Computer Society
0272-1716/13/$31.00 © 2013 IEEE
(a)
(b) Figure 1. Nobuyuki Umetani and his colleagues’ guided furniture design system. (a) The system can determine that an initial design (see the left image) is physically unstable and suggest alterations to satisfy robustness criteria.1 (b) A simple chair design validates the system. (Images © Nobuyuki Umetani; used with permission.)
(a)
(b)
Figure 2. The Shapeways Creator tools. (a) The Cufflink Creator converts letters into a 3D object. (b) The Ring Creator takes a bitmap image (see the inset) as input and generates a corresponding 3D model that can be 3D-printed. (Figure 2a © Shapeways; used with permission.)
Design-to-fabricate capabilities are also more frequently appearing in commercial systems, particularly as simple Web-based and tablet or mobile applications. For example, the Shapeways Creator tools (www.shapeways.com/create) let unskilled users create customized 3D models for printing. With the Cufflink Creator, users need only type in their initials and pick a font (see Figure 2a). The Ring Creator takes a gray-scale image as input, wraps it around a ring template, and extrudes a ring (see Figure 2b). This workflow lets novices use familiar image-editing tools to create unique 3D objects.
Other recent tools include Cookie Caster (http://cookiecaster.com), for designing custom cookie cutters (see Figure 3), and Crayon Creatures (http://crayoncreatures.com), a Web service that converts children’s drawings into 3D prints (see Figure 4). With many of these tools, a service bureau 3D-prints the object and mails it to the user. Laser cutting is also becoming increasingly available. Software such as Autodesk 123D Make (www.123dapp.com/make) lets unskilled users convert arbitrary 3D models to laser-cut fabrication plans with the push of a button (see Figure 5). As the library of freely available and easy-to-use IEEE Computer Graphics and Applications
27
Applications
(a)
(b)
Figure 3. Cookie Caster. (a) We designed a closed 2D polygon using a Web-based curve editor. (b) The software then automatically applied standard design rules for cookie cutters to create a 3D model suitable for printing.
(a)
design tools grows, users are combining them in interesting ways. Figure 6 shows an example from a recent Valentine’s Day event in Toronto called the 3D Printing Kissing Booth, created by the design studio HotPopFactory (www.hotpopfactory. com). The booth used a Microsoft Kinect to 3Dscan participants. A quick design session turned the scan into a custom, printable design, which the booth 3D-printed.
Meshmixer: Accessible General-Purpose Design Most design-to-fabricate tools closely heed the maxim “do one thing, and do it well” and hence
(b)
Figure 4. Taking (a) a child’s drawing, the Crayon Creatures Web service lets users easily inflate it in the third dimension and create (b) a color 3D print of the result. (Images © Crayon Creatures; used with permission.)
(a)
(b)
(c)
Figure 5. Autodesk 123D Make. (a) Users create a 3D model. (b) 123D Make generates a slicing pattern. (c) On the basis of the pattern, objects can be inexpensively fabricated at large scales using laser-cut cardboard slices. (Images © Autodesk; used with permission.) 28
November/December 2013
focus on solving a single design problem. This approach is particularly effective when inexperienced 3D designers are the target audience. However, in our experience, novices will inevitably (and often quickly) master any given task-specific tool and begin demanding more expressive capability. Unfortunately, scaling these simple tools to more complex design tasks is difficult. Many online discussions begin with the question “How do I do X?” and end with the suggestion that the user look into Blender or SolidWorks, extraordinarily complex professional tools. Currently, the chasm between these simple taskoriented fabrication tools and general-purpose 3D-design tools is simply too large, requiring significant learning time to cross. So, we’re developing and exploring tools that support a range of 3D-design tasks while providing high accessibility and utility to users of varying skill levels. In particular, Ryan Schmidt for several years has been developing meshmixer, and together we’ve explored using it in fabrication-related contexts. Meshmixer is available for free download for Window and OS X, at www.meshmixer.com. It includes tools for easy 3D mash-ups, virtual sculpting, and mesh repair, alongside more traditional shape-modeling functions such as extrusions and 3D transformations. An underlying motivation for bringing this disparate toolset together in a single interface is to simplify workflows. In particular, we want to make life easier for those who aren’t professional 3D artists, such as many members of the growing maker community. For this user base, the 3D design is a means to an end rather than the end itself, and the tools should take this into account. In the meshmixer project, we’re reimagining how a CAD tool should work at all levels, from the user interface to the fundamental data structures. For example, many 3D tools’ usability is compromised to varying degrees by limitations of the underlying mathematical shape representations—for example, Nurbs (nonuniform rational basis spline) or SuBD (subdivision) surfaces. These surfaces have desirable properties such as intrinsic smoothness and compact data structures with minimal degrees of freedom. However, they achieve these properties via 3D-structure constraints that users must explicitly manage. These constraints are opaque to novice users and lead to much frustration even for highly skilled users. Spurred by much of the recent computer graphics research in geometry processing, with meshmixer we discarded these structured shape representations in favor of a more flexible unstructured high-resolution triangle mesh. Doing this involved
(a)
(b)
Figure 6. The 3D Printing Kissing Booth. (a) A Microsoft Kinect created a 3D scan of two participants. (b) The booth quickly processed the scan and combined it with a base to create a unique 3D-printed keepsake. (Images © HotPopFactory; used with permission.)
trade-offs that leave meshmixer somewhat unsuitable for uses such as high-end film production or precision engineering. For example, with high-resolution triangle meshes, surface complexity is essentially free, whereas we can achieve smoothness only by applying computationally intensive variational mesh fairing. It remains to be seen whether this approach will lead to a CAD tool that’s viable for industrial design and engineering. However, we’re confident that our mesh-based approach greatly increases utility—and usability—for people who primarily want to make something awesome. As a result, we routinely find 3D-design novices using meshmixer to perform a variety of tasks. Figure 7 illustrates one interesting workflow in which meshmixer plays a critical role. A number of museums and galleries have hosted “scanathons,” after-hours events at which makers can create 3D scans of the collections using photogrammetry software such as Autodesk 123D Catch (www.123dapp.com/catch). This process often creates meshes containing many artifacts that prevent them from being 3D-printed; users have turned to meshmixer to fix this problem. Many users have posted the results to online model archives such as Thingiverse (http://thingiverse.com), where others can download and edit them—for example, to create mash-ups as in Figure 7. To demonstrate some of the meshmixer tools, we took the fish scan in Figure 7 and created a key chain (see Figure 8). To work with high-resolution meshes, the selection tool employs a brushing metaphor, much as in image-editing software. With this approach, we could quickly select and delete the unwanted areas of the surface and use automatic hole filling and mesh fairing to create an aesthetically pleasing watertight solid. We then simply dropped in a ring connector from our parts library and positioned it using a drag-and-drop composition tool. We applied variational mesh fairing to convert the resulting sharp transition IEEE Computer Graphics and Applications
29
Applications
(a)
(b)
Figure 7. Using meshmixer. (a) This software plays a role in many scan-clean-print workflows, which postprocess a coarse photogrammetric reconstruction—in this case, created with a cell phone camera—to convert the shell into a printable solid. (b) Many such objects have been posted online and incorporated by others into interesting mashups. (Figure 7b © Brian Evans, 2012; used with permission.)
(d) (a)
(b)
(c)
(e)
(f)
Figure 8. Our initial fish mesh wasn’t a closed volume and contained part of the background surface. (a) Using a triangle-selection brush, we quickly removed the unwanted area and filled the (b) large, complex hole with (c) new triangles. (d) We then dragged a ring onto the model and stitched it into the mesh along a small hole on the ring surface (not visible). So, the surface remained a single, watertight connected component. (e) Finally, we applied variational mesh fairing around this stitched transition to blend the ring into the fish, and (f) sent it to the 3D printer.
into a smooth one, similar to a traditional fillet. During these operations, the surface remained manifold and watertight; this let us send it directly to the 3D printer. The drag-and-drop tool we used for the example in Figure 8 is a good example of the style of interface we believe is suitable for inexperienced makers. With this tool, users can add arbitrarily complex parts to the mesh with a single click-anddrag. Novices find this satisfying because they can create detailed models that would otherwise be beyond their technical or artistic capabilities. Users can also easily share interesting parts. 30
November/December 2013
In a traditional CAD system, you might perform a similar operation using a Boolean union. However, that involves unconstrained 3D transformation—a difficult task for novices—and assumes that users understand technical jargon such as “Boolean union.” By wrapping multiple steps in a directmanipulation interface, with a live preview of the result, meshmixer simplifies what would otherwise be a complex multistep transaction. Of course, complex underlying geometric algorithms enable our drag-and-drop interaction’s simplicity.3 These algorithms make many decisions for the user and ultimately result in less control than an
Figure 9. At a workshop, girls aged 9 to 13 designed simple keychain and pendant objects in meshmixer and 3D-printed them. (Images © Ladies Learning Code; used with permission.)
expert might claim to want (although so far, experts seem to like this tool too). We strive for this level of interactive control in many of the meshmixer tools. In the best case, the many parameters of the underlying geometry-processing algorithms are abstracted into a simple, intuitive interface, which users interactively guide toward a desirable result.
Design-to-Print Workshops Like most research systems, meshmixer has minimal documentation and training materials and a relatively spartan visual design. However, we’ve iteratively refined the interaction design on the basis of several years of feedback from skilled users. To more formally explore how novices respond to meshmixer, we developed and taught a design-toprint workshop. We gave a one-hour introduction to meshmixer and 3D printing. Next, the participants spent an hour designing an object and 3D-printed the result. The workshop focused on jewelry customization—we provided various basic models of pendants, which the participants personalized. Constraining the task and base models helped keep the designed objects small and flat, ensuring reasonable print times. Many participants had never used a 3D UI before, so our instruction had to cover basic concepts such as what a 3D model is and how to manipulate a virtual 3D camera. We focused on meshmixer’s drag-and-drop composition tool because it let the participants easily add geometric details to a model from a parts library we provided. We also walked the participants through standard mesh selection and surface push-and-pull tools, and demonstrated the basics of meshmixer’s 3Dsculpting tools. Because design time was limited, we provided cheat sheets for the meshmixer UI and workflows, and “meshmixer mentors” were available to deal with software issues and questions. Our participants had no 3D-design experience, and, on the basis of our observations of novices
using other 3D tools, we had low expectations. In past experiments, novices had struggled with even basic camera manipulation. Subjects with no strong personal interest in 3D design are generally unwilling to work to overcome the inevitable barriers. This is particularly true when what they see on the screen in no way resembles the shape they’re trying to express. Given this previous experience, we were stunned by how quickly the participants learned to use meshmixer. Although many of the designs were relatively simple, all the subjects we talked to clearly enjoyed the experience. Several of them painstakingly created detailed 3D designs (which the consumer-level 3D printer’s limited resolution unfortunately didn’t always capture). This workshop’s success resulted in an ongoing series based on our materials (available at http://meshmixer.com/help/index.html). Figure 9 shows a workshop (taught by other instructors) in which girls aged 9 to 13 participated. It seems unlikely that such an event minus the 3D printing would have been remotely as successful. We strongly believe that because the participants were designing a real object—something they were going to 3D-print and take home—they were much more motivated to learn to use meshmixer. We’ve had similar experiences on a smaller scale with consumer-level 3D printers in the office, where our coworkers’ initial response has often been an excited “Can I make something?” These experiences suggest that digital fabrication might provide a strong incentive for a level of basic design literacy that previously didn’t exist.
Fabrication Guidance for 3D Printing The feedback and observations we’ve gathered from the workshops have led to a number of usability improvements in meshmixer. But perhaps a more significant by-product was an awareness of just how messy the digital-to-physical transition still is. Although 3D printing has an aspect of IEEE Computer Graphics and Applications
31
Applications
(a)
(b)
(c)
Figure 10. Dealing with overhangs. (a) If a model has too shallow a draft angle (the top image) or regions that are completely unsupported from below (the bottom image), drooping will result. (b) Visualizing surface regions that violate draft angle restrictions (the top image) lets users modify the shape slightly to improve printability (the bottom image). (c) Alternatively, we can automatically generate snap-off support structures to mitigate drooping.
magic, the reality is that the machines have many limitations that aren’t immediately obvious. So, a little bit of care taken in the digital stage often can lead to vastly improved physical objects.
proach scales with the user’s skill level. Novices can simply autogenerate support geometry where needed, whereas experts can combine their design knowledge with the interactive feedback to optimize an object for 3D printing.
Overhangs For example, consumer-level single-material fuseddeposition-modeling printers (such as the popular MakerBots) print in layers from bottom to top, essentially by extruding a tube of molten plastic. For this to work, something must be supporting each layer; otherwise, the plastic will fall into empty space. The rule of thumb is that parts of the model with draft angles under 45 degrees (overhangs) will droop when printing (see Figure 10). This is a significant design constraint that can be mitigated by printing support structures. Ideally, these structures snap off easily after printing; in practice, they often noticeably degrade print quality. Minimizing overhangs is a complex design constraint that novices have difficulty taking into account. Even skilled designers have problems determining visually where overhangs exist on complex free-form surfaces. Meshmixer includes a tool to identify overhangs and visualize them via a surface shader (see Figure 10). For overhangs for which small changes to the shape are acceptable, users can often use sculpting brushes to slightly tweak the surface to correct the draft angle. Or, meshmixer can automatically generate support posts for a particular overhang. This guided ap32
November/December 2013
Orientation Issues Besides determining the total overhang area, the model’s spatial orientation can have implications for model robustness. Between-layer bonds are weaker than in-layer bonds, so thin parts printed vertically are fragile. As a result, something as simple as changing the model’s orientation relative to the print bed can vastly affect the print strength. Some tools automatically determine an orientation that minimizes the total area requiring support, but this can result in support interfaces (which lower surface quality) covering important parts of the model. We’re exploring support minimization techniques that maximize part strength and surface quality for areas users deem the most important. Another important factor is whether an object will be stable in its intended orientation. Figure 11 shows screenshots from an interface that tells users whether a printed object will fall over. This is a straightforward geometric computation—an object is stable if its center of mass, when projected to the ground plane, lies within the convex hull of its ground contact points. Visualizations of the center of mass and its projected position relative to the convex hull support analysis of the object’s stabil-
(a)
(b)
Figure 11. Creating stable models. (a) A straightforward geometric analysis lets us determine whether this robot model will stand up under gravity. The red ball indicates that it won’t. (b) The color map indicates the total surface area that will require support structures if the print bed is aligned with the corresponding normal direction.
ity. Recently, Romain Prévost and his colleagues described techniques to automatically optimize an object to maximize stability.4
Tool Adaptability Many of the exciting applications of 3D printing— fixing broken parts, printing entire functional assemblies, and so on—involve tolerances that might depend on not only the printing technology or particular printer model but also machine-specific factors and even ambient effects such as room temperature. Our design tools must be material- and process-aware, to guide users through the complexities of adapting a design to different manufacturing techniques. In this direction, researchers are exploring interfaces that analyze a virtual design’s physical properties5 and automatically optimize the model to improve printability6 or interactively guide users toward a more desirable result.1
Improving the Interface These examples illustrate the complexity of helping novices design for fabrication in more general contexts, in which the interface can’t constrain users to designs that are safe for 3D printing. However, we’ve only scratched the surface of what might be useful in a design tool such as meshmixer. For example, although visualizations of geometricanalysis and simulation results are helpful, novice users likely won’t understand how to interpret the feedback without instruction or know how to resolve tricky fabrication issues. So how do we improve the interface? More informative visual guidance will be necessary to enhance usability, whereas suggestions and automatic changes will help users improve their de
signs. We’ll need both approaches to really unlock digital fabrication’s power, creating new design tools—and entirely new design approaches—by combining advances in user interfaces, geometry processing, and machine learning.
D
igital fabrication is an active, exciting space, with new and interesting problems whose solutions will directly impact industry and society in ways we can’t imagine. Although we’re wary of the hype around 3D printing, we also note that each 3D-printing workshop we’ve observed in Toronto has been larger—and has sold out more quickly— than the previous one. As the price for entry-level 3D printers drops under US$500, the critical question is, what will people do with them? Regardless of the task, interactive 3D-design tools will clearly play a key role. Our goal is a system that can guide any user— from hobbyist to professional—through creating a design suitable for direct fabrication. A key design criterion we didn’t cover in this article is structural integrity, both during manufacturing and afterward, when the object is put to its intended use. Current research treats this criterion mainly as part of postprocessing.5.6 However, in our opinion, the design cycle should integrate such analysis, running in the background and providing interactive guidance. Another critical and complex issue is tolerances. It’s all too easy to design a part that’s too thin to print properly or an assembly that won’t fit together when 3D-printed. Tolerances in digital fabrication often have complexities; for example, in some sintering processes, the tolerance for a IEEE Computer Graphics and Applications
33
Applications
small hole depends on local thickness. Similarly, we also must consider cost and material use. New approaches to shape analysis will likely be needed to guide users through the gauntlet of machinedependent parameters. New tools for digital fabrication have important intermediate-term ramifications. Get the tool chain right—easy to learn, scalable in complexity, and accessible—and digital fabrication could spur widespread engagement in creative making. Get it wrong—hard to learn, limited in capability, and focused on restricted social groups—and digital fabrication could become just another mechanism for delivering products to consumers.
Interactive Interface for Surface Composition, tech. report CSRG-611, Dept. of Computer Science, Univ. of Toronto, 2010. 4. R. Prévost et al., “Make It Stand: Balancing Shapes for 3D Fabrication,” ACM Trans. Graphics, vol. 32, no. 4, 2013, article 81; http://igl.ethz.ch/projects/ make-it-stand. 5. Q. Zhou, J. Panetta, and D. Zorin, “Worst-Case Structural Analysis,” ACM Trans. Graphics, vol. 32, no. 4, 2013, article 137; www.cs.nyu.edu/~qnzhou/ projects/StructAys. 6. O. Stava et al., “Stress Relief: Improving Structural Strength of 3D Printable Objects,” ACM Trans. Graphics, vol. 31, no. 4, 2012, article 48; http://hpcg. purdue.edu/?page=publication&id=164.
References
Ryan Schmidt is a research scientist at Autodesk Research. Contact him at [email protected].
1. N. Umetani, T. Igarashi, and N. Mitra, “Guided Exploration of Physically Valid Shapes for Furniture Design,” ACM Trans. Graphics, vol. 31, no. 4, 2012, article 86; www-ui.is.s.u-tokyo. ac.jp/~ume/GuidedExploration/2012_siggraph_ GuidedExploration.html. 2. A. Rivers, A. Adams, and F. Durand, “Sculpting by Numbers,” ACM Trans. Graphics, vol. 31, no. 6, 2012, article 157; www.alecrivers.com/sculptingbynumbers. 3. R. Schmidt and K. Singh, Drag, Drop, and Clone: An
Matt Ratto is an assistant professor and the director of the Semaphore Research Cluster on Mobile and Pervasive Computing and the Critical Making Lab in the University of Toronto’s Faculty of Information. Contact him at matt. [email protected]. Contact department editor Mike Potel at potel@wildcrest. com.
Take the CS Library wherever you go! IEEE Computer Society magazines and Transactions are now available to subscribers in the portable ePub format. Just download the articles from the IEEE Computer Society Digital Library, and you can read them on any device that supports ePub. For more information, including a list of compatible devices, visit
www.computer.org/epub
34
November/December 2013
Editor: Mike Potel
Design-to-Fabricate Maker Hardware Requires Maker Software Ryan Schmidt Autodesk Research Matt Ratto University of Toronto
C
omputing’s evolution is driven largely by the development of novel hardware capabilities, which can be fully exploited only by new types of software interfaces. This was particularly evident in digital fabrication’s early days, when the development of CNC (computer numerical control) machines at MIT led to an interest in CAD tools, essentially to “create content” for this new hardware. Ivan Sutherland’s seminal Sketchpad system—and hence the first GUI—was one fruit of this endeavor. Recent years have seen a significant leap forward in rapid-prototyping hardware, particularly in 3D printing. At the high end, 3D printers have micro meter resolution and can combine flexible, rigid, and transparent materials. Consumer-level systems have output quality similar to machines that cost tens of thousands of dollars just a few years ago. And mailorder 3D-printing houses provide affordable access to a range of materials, from metals to ceramics. Now that 3D-printing hardware has become vastly more accessible, there’s a growing community of makers looking for software that lets them create content for their new devices. Of course, such tools already exist, in the form of professional CAD software. However, much like early digital-fabrication hardware, these tools’ cost and complexity are beyond the hobbyist maker’s reach. Here, we take a look at how researchers have tried to overcome this problem by creating user-friendly software that assists the design of 3D fabricated objects. In particular, we focus on the Autodesk meshmixer toolset.
CAD Tools for Makers Traditional CAD packages aim to support the pipelines and practices of professional engineering and 26
November/December 2013
industrial design. In the face of these rigid processes, innovation in easy-to-use CAD tools largely ground to a halt outside of academic research. However, the democratization we’re seeing in digital-fabrication hardware brings with it a much larger community interested in designing their own 3D objects. Even if they’re professionals by day, when they put on the maker hat, they have the freedom and desire to explore novel, unconventional tools. So, digital fabrication provides an ideal playground for researchers and software developers—the tool makers—to experiment with new approaches to 2D and 3D design. This is spurring the development of new types of design tools, particularly those that deeply integrate the knowledge that the end goal is a physical object. Although academic computer graphics researchers have always been interested in easy-to-use design tools, research is increasingly incorporating fabrication constraints into the design process. These design-to-fabricate systems enable the rapid creation of complex physically realizable designs that would otherwise be well beyond even moderately skilled users’ capabilities. A notable example is Nobuyuki Umetani and his colleagues’ guided furniture design tool (see Figure 1).1 Users can, for instance, explore any number of wildly complex bookshelf designs, with the system providing subtle hints to guide a design toward one that’s physically stable and manufacturable. Some research has taken the reverse approach, applying geometric analysis to augment users’ realworld fabrication abilities. In the Sculpting-byNumbers project, Alec Rivers and his colleagues helped novices create clay sculptures of given 3D models.2 Using a camera and projector, the system analyzed the sculpture and displayed guidance information directly on it.
Published by the IEEE Computer Society
0272-1716/13/$31.00 © 2013 IEEE
(a)
(b) Figure 1. Nobuyuki Umetani and his colleagues’ guided furniture design system. (a) The system can determine that an initial design (see the left image) is physically unstable and suggest alterations to satisfy robustness criteria.1 (b) A simple chair design validates the system. (Images © Nobuyuki Umetani; used with permission.)
(a)
(b)
Figure 2. The Shapeways Creator tools. (a) The Cufflink Creator converts letters into a 3D object. (b) The Ring Creator takes a bitmap image (see the inset) as input and generates a corresponding 3D model that can be 3D-printed. (Figure 2a © Shapeways; used with permission.)
Design-to-fabricate capabilities are also more frequently appearing in commercial systems, particularly as simple Web-based and tablet or mobile applications. For example, the Shapeways Creator tools (www.shapeways.com/create) let unskilled users create customized 3D models for printing. With the Cufflink Creator, users need only type in their initials and pick a font (see Figure 2a). The Ring Creator takes a gray-scale image as input, wraps it around a ring template, and extrudes a ring (see Figure 2b). This workflow lets novices use familiar image-editing tools to create unique 3D objects.
Other recent tools include Cookie Caster (http://cookiecaster.com), for designing custom cookie cutters (see Figure 3), and Crayon Creatures (http://crayoncreatures.com), a Web service that converts children’s drawings into 3D prints (see Figure 4). With many of these tools, a service bureau 3D-prints the object and mails it to the user. Laser cutting is also becoming increasingly available. Software such as Autodesk 123D Make (www.123dapp.com/make) lets unskilled users convert arbitrary 3D models to laser-cut fabrication plans with the push of a button (see Figure 5). As the library of freely available and easy-to-use IEEE Computer Graphics and Applications
27
Applications
(a)
(b)
Figure 3. Cookie Caster. (a) We designed a closed 2D polygon using a Web-based curve editor. (b) The software then automatically applied standard design rules for cookie cutters to create a 3D model suitable for printing.
(a)
design tools grows, users are combining them in interesting ways. Figure 6 shows an example from a recent Valentine’s Day event in Toronto called the 3D Printing Kissing Booth, created by the design studio HotPopFactory (www.hotpopfactory. com). The booth used a Microsoft Kinect to 3Dscan participants. A quick design session turned the scan into a custom, printable design, which the booth 3D-printed.
Meshmixer: Accessible General-Purpose Design Most design-to-fabricate tools closely heed the maxim “do one thing, and do it well” and hence
(b)
Figure 4. Taking (a) a child’s drawing, the Crayon Creatures Web service lets users easily inflate it in the third dimension and create (b) a color 3D print of the result. (Images © Crayon Creatures; used with permission.)
(a)
(b)
(c)
Figure 5. Autodesk 123D Make. (a) Users create a 3D model. (b) 123D Make generates a slicing pattern. (c) On the basis of the pattern, objects can be inexpensively fabricated at large scales using laser-cut cardboard slices. (Images © Autodesk; used with permission.) 28
November/December 2013
focus on solving a single design problem. This approach is particularly effective when inexperienced 3D designers are the target audience. However, in our experience, novices will inevitably (and often quickly) master any given task-specific tool and begin demanding more expressive capability. Unfortunately, scaling these simple tools to more complex design tasks is difficult. Many online discussions begin with the question “How do I do X?” and end with the suggestion that the user look into Blender or SolidWorks, extraordinarily complex professional tools. Currently, the chasm between these simple taskoriented fabrication tools and general-purpose 3D-design tools is simply too large, requiring significant learning time to cross. So, we’re developing and exploring tools that support a range of 3D-design tasks while providing high accessibility and utility to users of varying skill levels. In particular, Ryan Schmidt for several years has been developing meshmixer, and together we’ve explored using it in fabrication-related contexts. Meshmixer is available for free download for Window and OS X, at www.meshmixer.com. It includes tools for easy 3D mash-ups, virtual sculpting, and mesh repair, alongside more traditional shape-modeling functions such as extrusions and 3D transformations. An underlying motivation for bringing this disparate toolset together in a single interface is to simplify workflows. In particular, we want to make life easier for those who aren’t professional 3D artists, such as many members of the growing maker community. For this user base, the 3D design is a means to an end rather than the end itself, and the tools should take this into account. In the meshmixer project, we’re reimagining how a CAD tool should work at all levels, from the user interface to the fundamental data structures. For example, many 3D tools’ usability is compromised to varying degrees by limitations of the underlying mathematical shape representations—for example, Nurbs (nonuniform rational basis spline) or SuBD (subdivision) surfaces. These surfaces have desirable properties such as intrinsic smoothness and compact data structures with minimal degrees of freedom. However, they achieve these properties via 3D-structure constraints that users must explicitly manage. These constraints are opaque to novice users and lead to much frustration even for highly skilled users. Spurred by much of the recent computer graphics research in geometry processing, with meshmixer we discarded these structured shape representations in favor of a more flexible unstructured high-resolution triangle mesh. Doing this involved
(a)
(b)
Figure 6. The 3D Printing Kissing Booth. (a) A Microsoft Kinect created a 3D scan of two participants. (b) The booth quickly processed the scan and combined it with a base to create a unique 3D-printed keepsake. (Images © HotPopFactory; used with permission.)
trade-offs that leave meshmixer somewhat unsuitable for uses such as high-end film production or precision engineering. For example, with high-resolution triangle meshes, surface complexity is essentially free, whereas we can achieve smoothness only by applying computationally intensive variational mesh fairing. It remains to be seen whether this approach will lead to a CAD tool that’s viable for industrial design and engineering. However, we’re confident that our mesh-based approach greatly increases utility—and usability—for people who primarily want to make something awesome. As a result, we routinely find 3D-design novices using meshmixer to perform a variety of tasks. Figure 7 illustrates one interesting workflow in which meshmixer plays a critical role. A number of museums and galleries have hosted “scanathons,” after-hours events at which makers can create 3D scans of the collections using photogrammetry software such as Autodesk 123D Catch (www.123dapp.com/catch). This process often creates meshes containing many artifacts that prevent them from being 3D-printed; users have turned to meshmixer to fix this problem. Many users have posted the results to online model archives such as Thingiverse (http://thingiverse.com), where others can download and edit them—for example, to create mash-ups as in Figure 7. To demonstrate some of the meshmixer tools, we took the fish scan in Figure 7 and created a key chain (see Figure 8). To work with high-resolution meshes, the selection tool employs a brushing metaphor, much as in image-editing software. With this approach, we could quickly select and delete the unwanted areas of the surface and use automatic hole filling and mesh fairing to create an aesthetically pleasing watertight solid. We then simply dropped in a ring connector from our parts library and positioned it using a drag-and-drop composition tool. We applied variational mesh fairing to convert the resulting sharp transition IEEE Computer Graphics and Applications
29
Applications
(a)
(b)
Figure 7. Using meshmixer. (a) This software plays a role in many scan-clean-print workflows, which postprocess a coarse photogrammetric reconstruction—in this case, created with a cell phone camera—to convert the shell into a printable solid. (b) Many such objects have been posted online and incorporated by others into interesting mashups. (Figure 7b © Brian Evans, 2012; used with permission.)
(d) (a)
(b)
(c)
(e)
(f)
Figure 8. Our initial fish mesh wasn’t a closed volume and contained part of the background surface. (a) Using a triangle-selection brush, we quickly removed the unwanted area and filled the (b) large, complex hole with (c) new triangles. (d) We then dragged a ring onto the model and stitched it into the mesh along a small hole on the ring surface (not visible). So, the surface remained a single, watertight connected component. (e) Finally, we applied variational mesh fairing around this stitched transition to blend the ring into the fish, and (f) sent it to the 3D printer.
into a smooth one, similar to a traditional fillet. During these operations, the surface remained manifold and watertight; this let us send it directly to the 3D printer. The drag-and-drop tool we used for the example in Figure 8 is a good example of the style of interface we believe is suitable for inexperienced makers. With this tool, users can add arbitrarily complex parts to the mesh with a single click-anddrag. Novices find this satisfying because they can create detailed models that would otherwise be beyond their technical or artistic capabilities. Users can also easily share interesting parts. 30
November/December 2013
In a traditional CAD system, you might perform a similar operation using a Boolean union. However, that involves unconstrained 3D transformation—a difficult task for novices—and assumes that users understand technical jargon such as “Boolean union.” By wrapping multiple steps in a directmanipulation interface, with a live preview of the result, meshmixer simplifies what would otherwise be a complex multistep transaction. Of course, complex underlying geometric algorithms enable our drag-and-drop interaction’s simplicity.3 These algorithms make many decisions for the user and ultimately result in less control than an
Figure 9. At a workshop, girls aged 9 to 13 designed simple keychain and pendant objects in meshmixer and 3D-printed them. (Images © Ladies Learning Code; used with permission.)
expert might claim to want (although so far, experts seem to like this tool too). We strive for this level of interactive control in many of the meshmixer tools. In the best case, the many parameters of the underlying geometry-processing algorithms are abstracted into a simple, intuitive interface, which users interactively guide toward a desirable result.
Design-to-Print Workshops Like most research systems, meshmixer has minimal documentation and training materials and a relatively spartan visual design. However, we’ve iteratively refined the interaction design on the basis of several years of feedback from skilled users. To more formally explore how novices respond to meshmixer, we developed and taught a design-toprint workshop. We gave a one-hour introduction to meshmixer and 3D printing. Next, the participants spent an hour designing an object and 3D-printed the result. The workshop focused on jewelry customization—we provided various basic models of pendants, which the participants personalized. Constraining the task and base models helped keep the designed objects small and flat, ensuring reasonable print times. Many participants had never used a 3D UI before, so our instruction had to cover basic concepts such as what a 3D model is and how to manipulate a virtual 3D camera. We focused on meshmixer’s drag-and-drop composition tool because it let the participants easily add geometric details to a model from a parts library we provided. We also walked the participants through standard mesh selection and surface push-and-pull tools, and demonstrated the basics of meshmixer’s 3Dsculpting tools. Because design time was limited, we provided cheat sheets for the meshmixer UI and workflows, and “meshmixer mentors” were available to deal with software issues and questions. Our participants had no 3D-design experience, and, on the basis of our observations of novices
using other 3D tools, we had low expectations. In past experiments, novices had struggled with even basic camera manipulation. Subjects with no strong personal interest in 3D design are generally unwilling to work to overcome the inevitable barriers. This is particularly true when what they see on the screen in no way resembles the shape they’re trying to express. Given this previous experience, we were stunned by how quickly the participants learned to use meshmixer. Although many of the designs were relatively simple, all the subjects we talked to clearly enjoyed the experience. Several of them painstakingly created detailed 3D designs (which the consumer-level 3D printer’s limited resolution unfortunately didn’t always capture). This workshop’s success resulted in an ongoing series based on our materials (available at http://meshmixer.com/help/index.html). Figure 9 shows a workshop (taught by other instructors) in which girls aged 9 to 13 participated. It seems unlikely that such an event minus the 3D printing would have been remotely as successful. We strongly believe that because the participants were designing a real object—something they were going to 3D-print and take home—they were much more motivated to learn to use meshmixer. We’ve had similar experiences on a smaller scale with consumer-level 3D printers in the office, where our coworkers’ initial response has often been an excited “Can I make something?” These experiences suggest that digital fabrication might provide a strong incentive for a level of basic design literacy that previously didn’t exist.
Fabrication Guidance for 3D Printing The feedback and observations we’ve gathered from the workshops have led to a number of usability improvements in meshmixer. But perhaps a more significant by-product was an awareness of just how messy the digital-to-physical transition still is. Although 3D printing has an aspect of IEEE Computer Graphics and Applications
31
Applications
(a)
(b)
(c)
Figure 10. Dealing with overhangs. (a) If a model has too shallow a draft angle (the top image) or regions that are completely unsupported from below (the bottom image), drooping will result. (b) Visualizing surface regions that violate draft angle restrictions (the top image) lets users modify the shape slightly to improve printability (the bottom image). (c) Alternatively, we can automatically generate snap-off support structures to mitigate drooping.
magic, the reality is that the machines have many limitations that aren’t immediately obvious. So, a little bit of care taken in the digital stage often can lead to vastly improved physical objects.
proach scales with the user’s skill level. Novices can simply autogenerate support geometry where needed, whereas experts can combine their design knowledge with the interactive feedback to optimize an object for 3D printing.
Overhangs For example, consumer-level single-material fuseddeposition-modeling printers (such as the popular MakerBots) print in layers from bottom to top, essentially by extruding a tube of molten plastic. For this to work, something must be supporting each layer; otherwise, the plastic will fall into empty space. The rule of thumb is that parts of the model with draft angles under 45 degrees (overhangs) will droop when printing (see Figure 10). This is a significant design constraint that can be mitigated by printing support structures. Ideally, these structures snap off easily after printing; in practice, they often noticeably degrade print quality. Minimizing overhangs is a complex design constraint that novices have difficulty taking into account. Even skilled designers have problems determining visually where overhangs exist on complex free-form surfaces. Meshmixer includes a tool to identify overhangs and visualize them via a surface shader (see Figure 10). For overhangs for which small changes to the shape are acceptable, users can often use sculpting brushes to slightly tweak the surface to correct the draft angle. Or, meshmixer can automatically generate support posts for a particular overhang. This guided ap32
November/December 2013
Orientation Issues Besides determining the total overhang area, the model’s spatial orientation can have implications for model robustness. Between-layer bonds are weaker than in-layer bonds, so thin parts printed vertically are fragile. As a result, something as simple as changing the model’s orientation relative to the print bed can vastly affect the print strength. Some tools automatically determine an orientation that minimizes the total area requiring support, but this can result in support interfaces (which lower surface quality) covering important parts of the model. We’re exploring support minimization techniques that maximize part strength and surface quality for areas users deem the most important. Another important factor is whether an object will be stable in its intended orientation. Figure 11 shows screenshots from an interface that tells users whether a printed object will fall over. This is a straightforward geometric computation—an object is stable if its center of mass, when projected to the ground plane, lies within the convex hull of its ground contact points. Visualizations of the center of mass and its projected position relative to the convex hull support analysis of the object’s stabil-
(a)
(b)
Figure 11. Creating stable models. (a) A straightforward geometric analysis lets us determine whether this robot model will stand up under gravity. The red ball indicates that it won’t. (b) The color map indicates the total surface area that will require support structures if the print bed is aligned with the corresponding normal direction.
ity. Recently, Romain Prévost and his colleagues described techniques to automatically optimize an object to maximize stability.4
Tool Adaptability Many of the exciting applications of 3D printing— fixing broken parts, printing entire functional assemblies, and so on—involve tolerances that might depend on not only the printing technology or particular printer model but also machine-specific factors and even ambient effects such as room temperature. Our design tools must be material- and process-aware, to guide users through the complexities of adapting a design to different manufacturing techniques. In this direction, researchers are exploring interfaces that analyze a virtual design’s physical properties5 and automatically optimize the model to improve printability6 or interactively guide users toward a more desirable result.1
Improving the Interface These examples illustrate the complexity of helping novices design for fabrication in more general contexts, in which the interface can’t constrain users to designs that are safe for 3D printing. However, we’ve only scratched the surface of what might be useful in a design tool such as meshmixer. For example, although visualizations of geometricanalysis and simulation results are helpful, novice users likely won’t understand how to interpret the feedback without instruction or know how to resolve tricky fabrication issues. So how do we improve the interface? More informative visual guidance will be necessary to enhance usability, whereas suggestions and automatic changes will help users improve their de
signs. We’ll need both approaches to really unlock digital fabrication’s power, creating new design tools—and entirely new design approaches—by combining advances in user interfaces, geometry processing, and machine learning.
D
igital fabrication is an active, exciting space, with new and interesting problems whose solutions will directly impact industry and society in ways we can’t imagine. Although we’re wary of the hype around 3D printing, we also note that each 3D-printing workshop we’ve observed in Toronto has been larger—and has sold out more quickly— than the previous one. As the price for entry-level 3D printers drops under US$500, the critical question is, what will people do with them? Regardless of the task, interactive 3D-design tools will clearly play a key role. Our goal is a system that can guide any user— from hobbyist to professional—through creating a design suitable for direct fabrication. A key design criterion we didn’t cover in this article is structural integrity, both during manufacturing and afterward, when the object is put to its intended use. Current research treats this criterion mainly as part of postprocessing.5.6 However, in our opinion, the design cycle should integrate such analysis, running in the background and providing interactive guidance. Another critical and complex issue is tolerances. It’s all too easy to design a part that’s too thin to print properly or an assembly that won’t fit together when 3D-printed. Tolerances in digital fabrication often have complexities; for example, in some sintering processes, the tolerance for a IEEE Computer Graphics and Applications
33
Applications
small hole depends on local thickness. Similarly, we also must consider cost and material use. New approaches to shape analysis will likely be needed to guide users through the gauntlet of machinedependent parameters. New tools for digital fabrication have important intermediate-term ramifications. Get the tool chain right—easy to learn, scalable in complexity, and accessible—and digital fabrication could spur widespread engagement in creative making. Get it wrong—hard to learn, limited in capability, and focused on restricted social groups—and digital fabrication could become just another mechanism for delivering products to consumers.
Interactive Interface for Surface Composition, tech. report CSRG-611, Dept. of Computer Science, Univ. of Toronto, 2010. 4. R. Prévost et al., “Make It Stand: Balancing Shapes for 3D Fabrication,” ACM Trans. Graphics, vol. 32, no. 4, 2013, article 81; http://igl.ethz.ch/projects/ make-it-stand. 5. Q. Zhou, J. Panetta, and D. Zorin, “Worst-Case Structural Analysis,” ACM Trans. Graphics, vol. 32, no. 4, 2013, article 137; www.cs.nyu.edu/~qnzhou/ projects/StructAys. 6. O. Stava et al., “Stress Relief: Improving Structural Strength of 3D Printable Objects,” ACM Trans. Graphics, vol. 31, no. 4, 2012, article 48; http://hpcg. purdue.edu/?page=publication&id=164.
References
Ryan Schmidt is a research scientist at Autodesk Research. Contact him at [email protected].
1. N. Umetani, T. Igarashi, and N. Mitra, “Guided Exploration of Physically Valid Shapes for Furniture Design,” ACM Trans. Graphics, vol. 31, no. 4, 2012, article 86; www-ui.is.s.u-tokyo. ac.jp/~ume/GuidedExploration/2012_siggraph_ GuidedExploration.html. 2. A. Rivers, A. Adams, and F. Durand, “Sculpting by Numbers,” ACM Trans. Graphics, vol. 31, no. 6, 2012, article 157; www.alecrivers.com/sculptingbynumbers. 3. R. Schmidt and K. Singh, Drag, Drop, and Clone: An
Matt Ratto is an assistant professor and the director of the Semaphore Research Cluster on Mobile and Pervasive Computing and the Critical Making Lab in the University of Toronto’s Faculty of Information. Contact him at matt. [email protected]. Contact department editor Mike Potel at potel@wildcrest. com.
Take the CS Library wherever you go! IEEE Computer Society magazines and Transactions are now available to subscribers in the portable ePub format. Just download the articles from the IEEE Computer Society Digital Library, and you can read them on any device that supports ePub. For more information, including a list of compatible devices, visit
www.computer.org/epub
34
November/December 2013
