|
Mark R. Cutkosky Dept. of Mechanical Engineering Stanford University Stanford, CA 94305 cutkosky@cdr.stanford.edu |
Jay M.
Tenenbaum
and Jay
Glicksman Enterprise Integration Technologies 800 El Camino Real Menlo Park, CA 94025 jmt@eit.com; jay@eit.com |
We describe an exercise in geographically distributed design and prototyping conducted by members of the ARPA MADE research community. The exercise tested emerging software and protocols for engineering collaboration over the Internet. The legacy of Madefast is an extensive project web that includes documentation about the design and the design process, as well as pointers to information sources, tools and services explored by the participants. Lessons from Madefast include the need for better methods for navigating and organizing on-line documentation, for coordinating and managing the efforts of cooperating groups, and for helping groups to assimilate the cultures of their virtual team-mates.
Similar challenges face defense contractors who need to respond rapidly to new requirements. The ARPA Manufacturing Automation and Design Engineering (MADE) program has been developing Internet-based tools, services, protocols and design methodologies that will allow contractors to compose teams of specialists from different locations and organizations as project needs arise. As a practical test of what the MADE program has achieved, members of the MADE community undertook an ambitious exercise in geographically distributed design and prototyping called Madefast. The exercise tested current MADE technology, identified future research directions and established a network of tools, services, documentation and contacts for the MADE community.
Madefast is an early example of a new and rapidly growing genre of projects that use the World Wide Web (WWW) extensively for collaborating and archiving results. Accordingly, the Web is the best primary resource for documenting the project. Our challenge in writing this paper is to linearize the presentation of the information contained in the web for publication in print. Readers with access to the Web are urged to read this paper while on-line, using it as their guide, and taking frequent excursions. The on-line version can be found at URL http://madefast.stanford.edu/ACM_paper.html. In the on-line version, references to documentation pages, tools and services are hyperlinked. In the printed version, we use endnotes to the URLs as surrogates for hyperlinks.
Following our overview of the project web and tools, we discuss some of the lessons learned from Madefast and consider extensions for the future.
Project overview
Madefast began as challenge issued at an ARPA program meeting in February
1994. Members of the ARPA MADE research community and program manager P.
Khosla resolved to design and prototype a defense-related product in six
months. It was agreed that the product should require a mix of
technologies and should take advantage of off-the-shelf components where
possible. A modified version of an optical tracker or seeker, such as used
in missiles and airborne surveillance, was suggested as a candidate
product.
Following the meeting, the project was launched with an electronic mail message broadcast to members of the MADE community outlining the proposed project and soliciting suggestions. Preliminary meetings and conversations by satellite videoconference, telephone and electronic mail established the project organization, with Stanford University and Enterprise Integration Technologies taking responsibility for project and web coordination and the University of Utah taking primary responsibility for mechanical design and fabrication of the (as yet undefined) Madefast product.
A project web was established during the first month of the project. The web included a chronology of meetings and milestones, a map of registered participants and top level pages for the design process and the design artifact.
Design specifications were released in an electronic mail message from ARPA on April 14, 1994. The requirement was to design a new prototype optical seeker, similar to those used in missiles and aircraft for tracking infrared targets. A seeker is complex system that combines optical, electronic and mechanical hardware as well as control software. As such it provides a good test of the ability of a diverse team of engineers to collaborate over a distance while using a variety of design and analysis tools. The Madefast seeker was given the non-lethal mission of tracking a dot of light from a hand-held laser pointer played against a wall.
The design effort began with contacts to defense contractors at Texas Instruments and Hughes who provided declassified drawings and expertise as a starting point for the redesign effort. These designs became a starting point for the design web. Although the Madefast seeker was not intended for launching, it was decided that the result would be more credible to the target audience in the defense industries if it came close to being launch-able. This decision entailed making the design fit inside a five-inch diameter tube and led to adoption of the same, compact mechanism used in military seekers in which mirror-based optics are mounted within gimbals for rotation about horizontal and vertical axes. A cut-away diagram of the design can be seen in Figure 2.
Preliminary designs and concepts were evaluated during the first few months of the project and subsystem fabrication and testing commenced over the summer of 1994. During this time, as the nature of the design became increasingly clear, other groups within the MADE community offered specific prototyping and analysis services for subsystems or components
A first prototype was built in September and demonstrated at an ARPA PI meeting in November, 1994. Subsequent work resulted in a second self-contained version of the device with on-board power and control electronics.
Figure 1: Madefast was a grass-roots effort involving industrial and academic participants from around the country. The on-line version of this figure is linked to the home page of each participating organization.
A tour of the Madefast web
The design of the documentation for Madefast was as challenging as the
design of the seeker itself. It was agreed early on to use the WWW as the
shared repository for all CAD models, notes, test results, calculations
and other information relating to the design. Because Madefast was a
community effort, a
"neutral" server
was established for top level
project pages. These pages
contain pointers to the webs maintained by participants at various sites;
it is in the participants' webs that the design resides.
It was also decided that parallel (but interconnected) webs should be maintained for documenting the Madefast project administration, design process, and design artifact. It is possible to navigate the Madefast web following any of these threads, often visiting the same pages.
Early attempts were also made to enforce a standard documentation style regarding the design, including decisions, tasks, rationale, etc. These attempts largely failed, for reasons that will be discussed in Section 7.1. Nonetheless, the level of documentation, while far short of the participants' ambitions, is better than typically found in design efforts involving a one-of-a-kind prototype designed by a hastily assembled team.
A good entry point for touring Madefast is the Project Overview page which describes what a seeker is and provides top-level pointers to the participants, project objectives, services and tools used, and design process. For an overall view of the design, the best top-level page is the Cu rrent State of the Design which contains pointers to information about the mechanical, optical and electronics subsystems which are described in pages maintained by the various participants. For example, Figure 2 shows the humped for the mechanical assemblies developed and fabricated by participants from the University of Utah.
A basic premise behind the documentation was that it is just as important to capture the process leading to a design as the design itself. This is especially true for a a distributed design exercise and for a project in which the team members have no history of prior collaboration and cannot readily foresee conflicts or bottlenecks in the design process. Consequently, a significant fraction of the Madefast web pages concern the design process and the rationale behind design choices. One such page summarizes the alternatives considered for the optical system. A diagram from this page is shown in Figure 3 and reveals that the final solution is just one of several considered. With a minor change in requirements (e.g., a larger diameter enclosure or a change in the tracking specifications) one of these other optical systems might be preferred for a redesign.
Other pages capture the chronology of meetings, demonstrations, and milestones. Again, this information can be useful for when revisiting a project for redesign. Where could time have been saved? What were the major sources of conflict?
Figure 2: Seeker cutaway view from the mechanical design page for the Madefast seeker.
Figure 3: A taxonomy of alternative optical systems considered.
Figure 4: The
The
Madefast Vision,
in which an engineer can use his or her own laptop
computer as an engineering notebook for personal and shared design
information and as a gateway to tools and services on the Internet.
Examples of engineering analysis tools and services provided by Madefast
participants and their associated URLs are listed in a
directory
of services and tools.
For example,
the preliminary optics analysis was performed with the aid of Rockwell's
Design Sheet and
final dynamics
analysis was performed using
SimLab from Cornell University. The seeker system was modeled using
DME from
the Stanford Knowledge Systems Laboratory,
which provided a symbolic model of the
seeker. The majority of the mechanical parts were designed and fabricated
using the
Alpha1 system from the
University of Utah. A composite tube and a light shroud were fabricated
using the rapid prototyping facilities at Michigan State University
and
Carnegie-Mellon University, respectively.
A concerted effort was made to exploit as many MADE tools and services as possible. Some of these resources are nearly commercial grade, while others are research prototypes. However, as discussed in the final section of this paper, many of the tools and services did enhance the richness and completeness of the on-line documentation even if they did not expedite the design process.
Collaboration and authoring tools
Participants also used several experimental tools to facilitate their
interactions with the Madefast Web and with each other. A multimedia
authoring environment helped them create and manage hyperlinked
engineering documents. A suite of collaboration tools enabled them to
share documents and ideas, both asynchronously via e-mail, and
synchronously in real time. These tools were integrated with the Web so
that downloading and launching them often involved nothing more than
pointing and clicking. The goal was to transform the WWW into a
collaboration medium for interdisciplinary engineering teams.
The Web Librarian is a model-based search utility for locating Web-documents. It is based on Dedal [Baudin et al. 1992], a system developed at NASA Ames for structuring and retrieving large amounts of design information. Like Dedal, the Web Librarian uses qualitative device models and heuristics to infer connections between queries and documents. For example, questions pertaining to ``design alternatives'' regarding the ``optical system'' lead to a list of links to web pages (with the best matches at the top of the list) containing discussions of candidate optical systems. Relevant documents can thus be found based on relationships and semantics, not just words that have been indexed. Users make queries through a WWW forms-based interface and receive links to potentially relevant documents. While the Web Librarian requires more work up front to create the models and links than to prepare a keyword index, that effort is repaid over the life cycle of a design and subsequent redesign efforts.
StoryBoard is a simple graphical editing environment that lets designers compose MIME multimedia messages, including graphics, animations, videos, and data files. These messages can be e-mailed to other designers or posted to the Web using HyperMail. StoryBoard provides two extensions to MIME, that have proved valuable for engineering collaboration. The first is a transparent overlay that allows graphical and textural annotations to be made without disturbing the integrity of the original message. The second is a new MIME content-type called pointer motion gestures. This extension enables users to ask and respond to questions about a design using synchronized recording and playback of mouse motion gestures accompanied by verbal annotation. For instance, an engineer can circle a dimension on a drawing while asking "Where did this value come from?"
Conference initiation and setup were handled non-intrusively using Mmphone. Mmphone is a MIME e-mail based rendezvous mechanism that is accessed through any web browser. A conference organizer completes an HTML form, checking off boxes corresponding to the collaboration capabilities he needs. The organizer also selects the desired participant from a pictorial index of Madefast personnel. A MIME message is then transmitted to the invitee, containing the information needed to enroll automatically in the conference.
The Madefast demonstration relied on point-to-point conferencing and Unix workstations running X. A reimplementation of SHAREd Web supporting IP Multicast and interoperation across Unix, PC and Mac platforms is under development. The architecture of the new system features a message bus on top of MBONE. This bus supports platform independence because the messages are application-specific, not platform-specific. An API (Application Programmer's Interface) makes it easy to write wrappers that make applications collaboration aware. The first application to be integrated is an MBONE version of Shared Mosaic, that can accommodate hundreds or even thousands of simultaneous users. The architecture scales well because only one copy of a web page is retrieved from the Web server, and that page is shared via Multicast.
Related Work
The Madefast exercise took advantage of several technologies and combined
them in interesting ways to carry out a complex design. In this section,
we describe related work in these different areas.
Distance collaboration using teleconferencing and shared media is an area of much research and development. Research prototypes include the Media Space project at Xerox Parc [Bly et al. 1993], the Cruiser [Fish et al. 1993] and Touring machine [Arango et al. 1993] projects at Bellcore, the Argo system at DEC [Gajewska et al. 1994], and the Ontario Telepresence Project [Buxton 1994]. These projects used proprietary systems and some analog video. In Madefast, we chis to use the MBONE tools [Eriksson 1994] because they are open and digital. This is expected to lead to systems that
Authoring to generate on-line engineer's notebooks is an important source of design rationale. Design notebooks have been developed on top of visual emacs [Lakin 1989] and Framemaker [Uejio 1991][Silva and Katz 1993]. WWWeasel and work being done at General Electric [Lewis et al. 1994] are now using the WWW as the basis for engineer's notebooks. This direction has immense potential since the WWW provides a powerful distribution mechanism.
Information retrieval is also an area of considerable interest. Many search engines exist for documents on the World Wide Web (e.g. WAIS, Personal Library Software, Verity and InfoSeek). Model-based retrieval techniques ([Weaver et al. 1989][Baudin et al. 1992][Celentano 1995]) are also being explored in the information retrieval research field. Web Librarian applies model-based retrieval techniques to the WWW in a document-oriented manner that brings these capabilities to a WWW-based search engine.
Discussion and future work
The Madefast exercise produced a working seeker prototype in record time.
But the more important legacy of Madefast is the living project web that
it has created. A project web such as the one created for Madefast is
useful for redesign, for the design of related projects, for engineering
education and for access to engineering tools, services and information in
the context of design examples. In the following paragraphs we expand on
these ideas.
Perhaps the single most important decision at the onset of the project was to make absolutely all documentation available on the World Wide Web. Since each participating group is responsible for its own web site, there is no central authority or distribution site. The main advantages of web-based design documentation are:
For similar reasons, the web was also useful for bringing new team members up to speed during the design and redesign projects. For example, it was useful to tell our control software consultants ``Here, go look at these web pages and send me a note when you've digested the material. Then we can talk.'' Comments from these consultants were also valuable for reorganizing the web to make it less confusing to newcomers.
A second use of the web is for learning about and revisiting tools and services used during a project. When revisiting an infrequently-used tool, engineers often prefer to go back to their notebooks to review how they used the tool on previous occasions, rather than diving straight into the generic documentation provided by the tool's creators. It is likely that the tool was previously used in a context similar to the present one and the extra contextual information, as well as any applications tips that were recorded in the earlier case, help the engineers to get reacquainted. By the same token, the Madefast web provides a convenient way to learn about tools and services available on-line.
Going a step further, we believe that a project web can provide a powerful combination of case-study and access to analysis tools for engineering education. The extra context of a working design example should help students to understand how the tools work, what they are good for, when they fail, and why.
The difficulty of organizing an evolving project hyperweb points to the need for tools to help us automate web organization and navigation as a by-product of a well-managed distributed design project. Existing tools for serving WWW pages treat the pages and their components as files. What is needed are tools that treat pages as documents with embedded media and that permit structuring of pages into units that can be manipulated as a group. Then, for example, pages could be reorganized from a chronological organization to an organization based on structural components of the device being designed, and all internal links would be automatically rewritten to match the new organization.
Advances in session creation and management are also needed, along with better procedures for capturing and reusing session information. To be used regularly, on-line teleconferencing tools must become as convenient as the competition: telephones, faxes and e-mail.
Other tools in the MADE community are equally sophisticated and demanding of experienced users. In many cases they were not representative of commercial grade software but were valuable services when combined with the expertise of the groups that knew them best. An interesting side issue is that the research programs were typically more open, and made more information explicit, than is the norm for commercial software. For example, the DME systems modeler from the Stanford Knowledge Systems Laboratory maintains explicit representations of constraints associated with every component. The resulting model provides richer documentation than a model in a commercial dynamic systems analysis program would typically provide.
Several advances are necessary before on-line services can compete with current practice. First, there is the issue of security. Companies will not send confidential information (design data, project information, or billing information) over the Internet to a service provider if they think it can be intercepted. Second, standards are important so that companies can use their in-house tools with any service provider. EDI standards for commerce are reaching a suitable state of maturity but are not well-supported on the Internet yet. Product standards, especially for advanced tools such as were used in Madefast, are still not sophisticated enough to handle all the data needed. Finally, interaction between users and services needs to be improved. The initial services being offered can be run as batch jobs with all the information provided before processing. Many interesting services, however, require interaction during the processing phase and user friendly interfaces are needed so that they can be used effectively.
The Madefast exercise validates the Internet as an information infrastructure for collaborative engineering. Although the results are encouraging, much work remains to make such an infrastructure commercially viable. For example, small and medium size providers of engineering and manufacturing services will need inexpensive starter kits to make their services accessible over the net. Designers will need directory services to help them locate these services and certification services to establish their qualifications. Research is needed to create the tools that will help organize on-line engineering information both for individuals and groups. ARPA's MADE program sponsors the development of such software and services.
Acknowledgments
The Madefast project was supported by ARPA under the MADE program. Special
thanks are due to G. Toye, L. Leifer and J. Wagner at Stanford and C.
Valiquette and S. Drake at Utah for the extensive time and effort they put
into Madefast. Many other individuals at the participant sites contributed
significantly to the Madefast project and while space does not permit us
to list them all here, we are sure that their contributions will become
apparent as readers explore the Madefast web.
[Baudin et al. 1992] Baudin, C., Gevins, J., Baya, V., and Mabogunje, A.,
``Dedal: Using Domain Concepts to Index Engineering Design Information,''
Proceedings of the 14th Annual Conference of the Cognitive Science
Society, August, 1992.
[Bly, S., Harrison, S., and Irwin, S., ``Media Spaces: Bringing People
Together in a Video, Audio, and Computing Environment,'' Communications of
the ACM, Vol. 36, No. 1, Jan. 1993, pp. 28-47.
[Buxton 1994] Buxton, W., ``Telepresence: Integrating Shared Task and
Person Spaces,'' Proceedings of Graphics Interface `92, Vancouver, B.C.,
May 1992.
[Celentano 1995] Celentano, A., Fugini, M.G., Pozzi, S., ``Knowledge-Based
Document Retrieval in Office Environments: The Kabiria System,'' ACM
Trans. on Information Systems, Vol. 13, No. 3, July 1995, pp. 237-268.
[Ellis et al. 1991] Ellis, C.A., Gibbs, S.J., and Rein, G.L., ``Groupware:
Some Issues and Experiences,'' Communications of the ACM, Vol. 34, No. 1,
January 1991, pp. 38-58.
[Eriksson 1994] Eriksson, H., ``MBONE: The Multicast Backbone,''
Communications of the ACM, Vol. 37, No. 8, August 1994, pp. 54-60.
[Fish et al. 1993] Fish, R.S., Kraut, R.E., Root, R.W., and Rice, R.E.,
``Video as a Technology for Information Communication,'' Communications of
the ACM, Vol. 36, No. 1, Jan. 1993, pp. 48-61.
[Gajewska et al. 1994] Gajewska, H., Kistler, J., Manasse, M.S., and
Redell, D.D., ``Argo: A System for Distributed Collaboration,'' Proc. ACM
Multimedia `94, San Francisco, October 1994, pp. 433-440.
[Glicksman et al. 1994] Glicksman, J., Kramer, G.A., and Mayer, N.P.,
"Internet Publishing via the World Wide Web", Proceedings Groupware '94,
San Jose CA, August 1994, pp. 431-442.
[Glicksman and Hart 1995] Glicksman, J. and Hart, C., ``A Web Librarian
for Structuring and Retrieval of Information, `` 2nd. Electronic
Documentation Workshop, Palo Alto, CA, May 1995, pp. 16-17.
[Karam, G.M., McLeod, B., and Boersma, G., ``A Network Services Interface
for Telepresence Applications,'' IEEE Intl. Symposium on Autonomous
Decentralized Systems, April 1995.
[Kumar et al. 1994] Kumar, V., Glicksman, J., and Kramer, G.A., ``A SHAREd
Web to Support Design Teams,'' Proc. Third Workshop on Enabling
Technologies: Infrastructure for Collaborative Enterprises, Morgantown,
West Virginia, April 1994, pp. 178-182.
[Lakin et al. 1989]
Lakin, F., Wambaugh, J. Leifer, L., Cannon, D., Sivard, C., ``The
electronic design notebook: performing medium and processing medium,''
Visual Computer: International Journal of Computer Graphics, Vol. 5, No.
4, 1989, pp. 214-226.
[Larner 1990] Larner, D.L., ``Factories, Objects, & Blackboards,'' AI
Expert, Vol. 5, No. 4, April 1990, pp. 38-45.
[Lewis et al. 1994] Lewis, J.W., Bernstein, B.M., Kenny, K.B., Mohr, G.,
Ostrowski, M.C., Sarachan, B.D., Sum, R.N., ``The Concurrent Engineering
Toolkit: A Network Agent for Manufacturing Cycle Time Reduction,''
Proceedings of CE `94: Concurrent Engineering Research and Applications,
Pittsburgh, PA, August 1994.
[Malone et al. 1995] Malone, T.W., Lai, K-Y., Fry, C., ``Experiments with
Oval: A `radically tailorable' tool for cooperative work,'' ACM Trans. on
Information Systems, Vol. 13, No. 2, April 1995, pp. 177-205.
[Pan et al. 1989] Pan, J.Y-C., Tenenbaum, J.M., Glicksman, J., ``A
Framework for Knowledge-Based Computer-Integrated Manufacturing,'' IEEE
Transactions on Semiconductor Manufacturing, Vol. SM-2, No. 2, May 1989,
pp. 33-46.
[Schmidt and Bannon 1992] Schmidt, K. and Bannon, L., ``Taking CSCW
Seriously: Supporting Articulation Work,'' Computer Supported Cooperative
Work, Vol. 1, No. 1-2, 1992, pp. 7-40.
[Silva and Katz 1993] Silva, M.J. and Katz, R.H., ``Active Documentation:
a New Interface for VLSI Design,'' Proceedings of the 30th Design
Automation Conference, Dallas Texas, June 1993, pp. 654-660.
[Toye et al. 1994] Toye, G., Cutkosky, M.R., Leifer, L.J., Tenenbaum,
J.M., and Glicksman, J., ``SHARE: A Methodology and Environment for
Collaborative Product Development,'' The International Journal of
Intelligent and Cooperative Information Systems , vol. 3, no. 2, June
1994, pp. 129-53.
[Uejio 1991] Uejio, W.H., ``Electronic Design Notebook for the DARPA
Initiative in Concurrent Engineering,'' CE Enabling Technology: Selected
Technical Papers, CERN, West Virginia University, April 23, 1991.
[Weaver et al. 1989] Weaver, M.T., France, R.K., Chen, Q., Fox, E.A.,
``Using a Frame-Based Language for Information Retrieval,'' Intl. Journal
of Intelligent Systems, Vol. 4, 1989, pp. 223-257.
[Williams and Lochovsky 1989] Williams, L.J. and Lochovsky, F.H.,
``Supporting Knowledge Migration in Organizations,'' Information
Processing 89, G.X. Ritter (ed.), August/September 1989, pp. 259-264.
References
[Arango et al. 1993] Arango, M. and 20 others, ``The Touring Machine
System,'' Communications of the ACM, Vol. 36, No. 1, Jan. 1993, pp.
68-77.