Summary of the Discussions at IWAR

The following Q&A summary was prepared from audio transcripts of the discussions at IWAR'98 by Gudrun Klinker, Reinhold Behringer, and David Mizell. A summary article on IWAR'98 was published in VR News [1] in Feb. 99

What is the cost of future AR systems?

The opinion was expressed that the major cost factor is in developing the technology, which is most likely to be driven by the cost for VR hardware. The application software only contributes to a minor share of the cost. Under certain circumstances, the cost of AR can actually be less than for a conventional information system, because the real environment and objects can be taken into account. AR eventually will be a commodity item.

What has to be done to build truly useful AR applications?

Several participants pointed out that in order to develop usable systems, more attention has to be paid to user issues by conducting user studies and by ``listening'' to potential end users. Tasks that can already be done well without AR, should not be in the center of research focus. Instead, tasks should be targeted towards applications where AR really can make a difference, e.g., design and prototyping.

A discrepancy of views was evident between academic and industrial researchers. Whereas the academic community is targeting the application of AR in daily life as a user-centric paradigm, implemented often in a kind of wearable system, industrial AR researchers argued that the current AR applications are strictly task specific.

In general, both sides agreed that technical progress in the display technology (latency, resolution) and in the general capabilities (repeatability, speed) is considered to be necessary for developing acceptable AR systems. AR systems should help eliminate human error and must provide means for doing tasks better than can be done with current technology. In order to give users convincing arguments for more advanced technology, not only the display aspect (3D displays) is important, but also the additional capabilities (e.g. networked information flow).

What will be a killer application of AR, and who will be the potential users?

This question was also controversial: One notion was that the AR applications will develop their greatest potential in factory environments, for military training and maintenance. Another view described AR applications as being most suitable for highly trained experts, such as air traffic controllers, video producers, biologists. A third notion of a possible user is the home consumer (toy and video market). The papers which were presented, actually covered all these types of AR applications.

A concrete ``killer app'' could not be defined; it was noted that so far only one AR system (Boeing's wire harness AR system for airplane manufacturing) is being produced and sold on a commercial basis.

What combinations of display/interaction technologies are promising?

The answers ranged from paradigms of imitating real-world interaction paradigms electronic paper and ink) to completely new ubiquitous display concepts. Portability and wearability is considered to be a huge leap forward, even if the precision of such an AR system would not be very high. Important are the interaction schemes provided to the user: keyboard-less, non-tethered methods (voice and gesture input) will provide intuitive means for exploring and controlling the information space accessible by the computer.

How will the currently used AR technology improve in 5-10 years?

This question got various answers, depending on the field of research of the individual. The ``computer vision community'' strongly believes in the role of computer vision (for registration and information capture). Another notion is that many more sensors will bring completely different applications of the computer technology, which currently is mostly used to perform ``paper-similar'' tasks. Possibly HMDs will have matured to light and comfortable devices which will not provoke the current objections.

HMDs -- necessary for AR?

This question was not asked explicitly, but many comments revealed the very ambivalent opinions of many AR researchers towards these visualization devices. Some consider HMDs as essential tools for providing immersive visual display, whereas others see them as bulky devices, not being accepted by users for more than very specific tasks. For many applications, handheld devices could also do the job of information display. Whether that can/should still be called {\em Augmented Reality}, is another question. Many researchers expressed the opinion that AR is more than just visualization: augmentation can be done for the aural sense as well as for other senses. And for the visual augmentation, other types of displays can provide task- and application-specific information. However, if HMDs will become lighter, smaller, and more comfortable to wear, they will be essential for user-centric registered visual augmentation.

Is there a common vision of AR?

During the discussions, a single vision of AR could not be deduced. Summarizing the statements which were made during the discussion, the AR community seems to be working in the direction of AR schemes that are wearable, ubiquitous, involving all human senses, and linking our environment with information. Defining AR by a certain technology is considered a limiting notion.

What are the issues for outdoor AR?

On the hardware side, outdoor applications must be light-weight and must run on low power. The display must be improved in terms of brightness, contrast, and of course weight. On the software side, registration in a non-constrained environment provides a large challenge. Computer vision, the method generally considered as most suitable for outdoor registration, must make more progress in order to replace/complement the conventional registration sensors. The tracking problem is far from being solved: especially in an outdoor scene a large variety of situations can occur which will cause current tracking systems loose track. Important for most visual tracking approaches is the knowledge of the 3D model of the tracked object or scene. Future work should focus on the automatic reconstruction of such 3D models and on incremental learning about the 3D environment.

Fiducial markers or natural features for CV-based tracking?

Many tracking systems rely on fiducial markers. This approach is acceptable by some industrial customers who can live with fiducials in an industrial environment; however, it is not adequate for use in outdoor scenarios where marking all objects just is not feasible. Therefore, progress is needed in using natural features instead of man-made fiducials for visual tracking. In order to compare the performance of various tracking systems, Mizell presented Klinker's idea of a ``tracker shoot-out'' in which a variety of tracker systems have to solve the task of tracking an object (CAD model provided).

Automated calibration or human-in-the-loop?

There were different opinions about the question, whether and to what extent users will be willing to spend time calibrating AR systems (esp. the trackers). If the calibration procedures take more time than the utilization of AR information presentation saves, the advantage of AR is questionable. Therefore, self-calibrating tracking systems are very compelling. However, the technology is not yet that mature, and a human may have to substitute for a lack of computer vision capability and manually calibrate the system -- at least until a better technical solution has been developed. Another question is if ``perfect'' registration is really necessary for all AR tasks. Since there is a cost function associated with registration precision, one might settle for less stringent requirements and still perform satisfactory AR.

How much do users have to be trained to use an AR system?

Specific tasks to be performed with an AR system actually can lessen the required training, as has been shown in the Boeing wire bundle AR project, where completely untrained users were able to work faster than trained personnel.

What is AR?

This question was actually also addressed in the previous discussion session, but restless minds again brought up discussions about this issue. The question was raised if AR is mostly integration. In fact, several techniques of various research fields are integrated into AR applications, such as computer vision, virtual reality, sound, AI, ergonomics; integration of these techniques is an important aspect of AR. However, AR has brought a few new research areas by itself, which are not part of any other existing research area, e.g., affine projection.

The specific requirements of AR in many areas (tracking, user-object/information interaction) trigger a lot of new research efforts in those ``integration'' fields, going far beyond their original scope. There was also some concern expressed that AR focuses too much on technical aspects, especially on registration, and neglects some fundamental human interface issues.

In answering the more general question for the definition of AR, Azuma referred to the definition in his survey (Azuma 1997), where AR is defined as: combining real and virtual, interactive in real-time, and registered in 3D. This definition was accepted by researchers doing work that fit within it, and rejected by those doing work which did not.

[1] Reinhold Behringer, Gudrun Klinker, and David Mizell. International Workshop on Augmented Reality. Edited by Mike Bevan. VR News, Jan./Feb. 1999, vol. 8, issue 1, pp. 18-20.
last update: Tuesday, January 06 2009, 11:56 AMMonday, February 13 2006, 06:27 PM (rb)