Stefanie Mueller: Interacting with Personal Fabrication Machines



Stefanie Mueller | X-Consortium Career Development Assistant Professor | Computer Science and Artificial Intelligence Lab

Personal fabrication tools, such as 3D printers, are on the way to enabling a future in which non-technical users will be able to create custom objects. With the recent drop in price for 3D printing hardware, these tools are about to enter the mass market: While the average consumer 3D printer was priced at $14,000 in 2007, today’s hardware costs, on average, only $500[10]. Given the decreasing price, it is not surprising that the number of sold consumer 3D printers has doubled every year[10].

While the hardware is now affordable and the number of people who own a 3D printer is increasing, only few users actually create new 3D models. Most download models from a platform, such as Thingiverse, and fabricate them on their 3D printers. At most, users adjust a few parameters of the model, such as changing its color or browsing between predetermined shape options.

I believe that personal fabrication has the potential for more: I envision a future in which, rather than just consuming existing content, 3D printers will allow non-technical users to create objects that only trained experts can create today. While there are many open challenges, I will use this article to discuss how we can improve the interaction model underlying current fabrication devices.

1.1 Interaction Model with 3D Printers Today

In the current interaction model, users sit at a computer and use a digital 3D editor to create a digital 3D model. Only at the end of the design process do users send the file to the 3D printer, which creates the object in one go. Because 3D printing is slow, this process tends to take hours of printing time for small objects and may even require overnight printing.

1.2 Drawing a Parallel to Personal Computing

Looking back in history, this interaction model with the delayed feedback was also common with early computers[1]. In the early ’60s, computers were so slow that the average program had to be executed overnight. Feedback was delayed until the next morning and if a program failed, users had to repeat the entire process, potentially waiting another night for results. Similar to 3D printers today, early computers were limited to expert users because when programs were executed in one go overnight, users had to know what they were doing to succeed.

1.3 Towards Turn-Taking and Direct Manipulation

However, today we are at a point at which even non-technical users can use personal computers. Beside many technical developments, two advances in the interaction model enabled this: (1) the move from executing in one go to turn-taking, and (2) the move from turn-taking to direct manipulation[3].

(1) Turn-taking: By decreasing the interaction unit to single requests, turn-taking systems, such as the command line, provided users with feedback after every input. This enabled the trial-and-error process that non-technical users tend to employ: quickly iterating through potential solutions and building each step onto the results of previous ones[2]. However, while the turn-taking interaction model provided a great step forward to making the technology available for non-technical users, the feedback cycle was still limited in that it consisted of two discrete steps: users first had to create an input and only afterwards received feedback.

(2) Direct manipulation: With the invention of direct manipulation[9] that further decreased the interaction unit to a single feature, users finally received real-time feedback: Input by the user and output by the system are so tightly coupled that no visible lag exists. This tightened feedback cycle has many benefits, among others that
“novices can learn basic functionality quickly” and “retain operational concepts”[8]. (See Figure 1.)

Description: ../../../Desktop/Screen%20Shot%202017-04-16%20at%208.36.45%20AM.

Figure 1. Server Feedback

As described above, the current interaction model of 3D printers requires objects to be fabricated in one go. Thus, from a human-computer interaction standpoint, we are today at the point at which we were with personal computers in the 1960s: Only few users are able to use the technology, and even for experts, it is a cumbersome process due to the delayed feedback.

1.4 Bringing Direct Manipulation to Fabrication

I argue that by repeating the evolution of the interaction model from personal computing, we will see the same benefits for personal fabrication: Direct manipulation will allow non-technical users to create physical objects as easily as they manipulate digital data with today’s personal computers.

A direct manipulation system for personal fabrication needs to have four main characteristics: (1) the physical environment is the workspace, not a digital editor; (2) users work hands-on on the physical workpiece through physical tools as known from traditional crafting; (3) each physical action results in immediate physical change, which can also be reversed; and (4) in contrast to traditional crafting, users receive support from a computer system that helps to achieve precision.

In the following section, we show examples of two systems that implement the requirements listed above and iteratively decrease the interaction unit from entire objects, to single elements, to features to achieve real-time physical feedback.

(1) Turn-taking: Interactive Laser-Cutting. To illustrate what a turn-taking system for personal fabrication might look like, we decrease the interaction unit from entire objects to single elements. In our system constructable[6], users draw with a laser pointer onto the workpiece inside a laser cutter. The drawing is captured with a camera. As soon as the user finishes drawing an element, such as a line, the constructable system beautifies the path and cuts it, resulting in physical output after every editing step. Different tools allow users to accomplish different tasks, such as copy-pasting physical shapes or creating matching finger joints between two edges. In addition, constructable ensures that all physical output is aligned (see Figure 2).

Description: ../../../Desktop/constructable.jpeg

Figure 2: constructable

While constructable allows for fast physical feedback, the interaction is still best described as turn-taking because it consists of two discrete steps: users first perform a command and then the system responds with physical feedback.

(2) Direct Manipulation: Continuous Forming. By decreasing the interaction unit even further to a single feature, we explore how to make the workpiece change while the user is manipulating it, resulting in real-time physical feedback: Input by the user and output by the fabrication device are so tightly coupled that no visible lag exists. Our system FormFab[7] provides such continuous physical feedback (see Figure 3). To accomplish this, FormFab neither adds nor subtracts material, but instead reshapes it (formative fabrication). A heat gun attached to a robotic arm warms up a thermoplastic sheet until it becomes compliant; users then control a pneumatic system that applies either pressure or vacuum, thereby pushing the material outwards or pulling it inwards. As users interact, they see the workpiece change continuously.

Description: ../../../Desktop/Screen%20Shot%202017-04-16%20at%208.35.38%20AM.

Figure 3. FormFab

1.5 Discussion

Direct manipulation systems for personal fabrication extend the range of problems novice users can tackle, but they are subject to the same limitations as those for personal computing: While they are useful for some design problems, they are less so for others. As Norman et al.[4] point out, direct manipulation interfaces are limited to operations that can be done on “visible objects” and have “difficulties handling variables” and “distinguishing an individual element from a representation of a set or class of elements.” Thus design problems that require more abstract thinking for which users must first sit down with a piece of paper first and make a detailed plan are better handled with traditional digital 3D editing. In addition, the systems presented in this article inherently scale 1:1 and do not offer a way of inspecting a detail in magnification, which limits users to projects that fit a particular scale. The same way that a saw and a wood chisel cannot replace a detailed design process, our systems cannot replace a complex 3D editing tool for trained engineers.

1.6 Outlook

We focused on using technology available today to explore interaction paradigms that will become possible in the future when fabrication actually gets faster. In recent years, we have begun rapidly advancing towards such a future. The recently introduced 3D printer Carbon3D, for instance, speeds up fabrication by up to 200 times.

While there is little data today that could prove a Moore’s law for 3D printers, an executive from 3D Systems, a leading manufacturer, noted in 2014 that 3D printing speed had, on average, doubled every 24 months over the previous 10 years[5]. If such a trend should materialize, it is not far-fetched to assume that fabrication technology will be able to provide feedback even for large high-resolution objects within seconds or event in real-time, thereby enabling a future in which digital displays will be replaced with physical displays that transform virtual reality into actual physical reality.


[1] P. Ceruzzi, A History of Modern Computing, 2nd edition. The MIT Press (2003).

[2] M. Csikszentmihalyi, Flow: The Psychology of Optimal Experience. Harper Perennial Modern Classics (1990).

[3] J. Grudin, A Moving Target: The Evolution of Human-Computer Interaction. Human-Computer Interaction Handbook, 3rd edition, Taylor and Francis (2012).

[4] E. Hutchins, J. Hollan, and D. Norman. Direct Manipulation Interfaces. Human-Computer-Interaction (1985), vol. 1, pp. 311-338.

[5] Moore’s law for 3D printing:

[6] S. Mueller, P. Lopes, P. Baudisch. Interactive Construction: Interactive Fabrication of Functional Mechanical Devices. Proceedings of the ACM UIST 2012, pp. 599-606.

[7] S. Mueller, A. Seufert, H. Peng, R. Kovacs, K. Reuss, T. Wollowski, F. Guimbetiere, P. Baudisch. Continuous Interactive Fabrication. Under review for ACM UIST 2017.

[8] B. Shneiderman. Direct Manipulation: A Step Beyond Programming Languages. Computer (1983), vol. 16, issue 8, pp. 57-69.

[9] B. Shneiderman. The Future of Interactive Systems and the Emergence of Direct Manipulation. Proceedings of the NYU Symposium on User Interfaces on Human Factors and Interactive Computer Systems (1984), pp. 1–28.

[10] Wohlers Report (2016).