Live Document

6: Solution Themes, Scientific Approaches, and Challenge Problems

One measure of the maturity of a research field is the emergence of a series of accepted practices and challenge problems that focus the attention of the field. Equally important is the identification of solution themes that cross applications. In this section, we survey several practices, challenge problems, and solution themes.

A. Accepted Practices
There are a number of accepted practices that are emerging in HRI. A key practice is to include experts from multiple disciplines on research efforts. These disciplines frequently include robotics, electrical and mechanical engineering, computer science, human-computer interaction, cognitive science, and human factors engineering. Other relevant disciplines include design, organizational behavior, and the social sciences. Importantly, the annual Human-Robot Interaction Conference encourages multi-disciplinary submissions and is establishing the practice of having all papers refereed by reviewers representing different disciplines bungee run for sale.

A second emerging practice is to create real systems (robot autonomy, interaction modes, and etc.) and then evaluate these systems using experiments with human subjects. Proof-of-concept technologies, although important, are less valuable than they would be if they were supported by careful experiments that identify key attributes of the design or principles that span applications. Identification of descriptive interaction phenomena is interesting, but elaboration on the psychological principles underlying these phenomena with an eye toward harnessing these principles in design is more useful. Thus, engineering, evaluation, and modeling are key aspects of HRI.

A third emerging practice is conducting experiments that include a careful blending of results from simulated and physical robots. On the one hand, because of cost and reliability issues, it is often difficult to conduct carefully controlled experiments with physical robots. On the other hand, it is often difficult to replicate simulation-only results with physical robots because the physical world presents challenges and details that are not present in many simulations. It is common to “embody” at least one portion of the interaction, be it a physical robot, some physical sensor, or real-world speech. Some research includes work using carefully controlled simulation environments and replication of selected results with physical robots. Others use wizard of Oz studies. Others form communities where roboticists design technologies and where other human factors researchers collect and analyze results from tests; this is the operating structure of part of the robot-assisted urban search and rescue (USAR) community [55]. Interestingly, at least one research group is exploring how a simulated user can help support the design of human-robot interfaces [295].

A fourth area of emerging effort is establishing standards and common metrics. The most complete survey of metrics is in [296], but much work on metrics exists in the literature including the proceedings of the annual PERMIS workshops. Standardization efforts have been strongest in the USAR domain [53, 232], but are also present in space applications and UAVs [284, 285].

A fifth emerging practice is the use of longitudinal studies. Such studies, which can last from several weeks to several months, require a considerable investment by researchers, both in terms of hours and financial resources. One reason that long-term studies are a relatively recent practice is that many robots were not reliable enough to work over the study period. The availability of reliable personal home robots and service robots in public areas has made such studies possible [130, 170]. The European COGNIRON project is a good example of a commitment to long-term studies [297]. Long-term studies shift research methodologies from carefully controlled small-scale experiments to other methodologies such as surveys and ethnography.

B. Challenge Problems
It is often useful to identify a set of challenge problems that focus the efforts of a community. HRI has a suite of challenge problems, some explicitly identified as such and others implicitly operating as such. In this section, we identify a collection of problems that are likely to shape HRI in the near future. For each problem, we discuss those aspects of the problem that make it particularly challenging and useful.

USAR is the most high profile of the HRI challenge problems. The attributes that define this as a challenge problem include the highly unstructured nature of USAR environments. This imposes strict challenges on robot mobility, communications, map-building, and situation awareness.

Developing robots to be used in military reconnaissance and combat is another high profile challenge area in HRI. Similar to USAR, environments tend to be unstructured, but perhaps more importantly operators may be required to operate under extreme stress in the presence of an adversary that is trying to prevent their success.

Space robotics is another area where the environment is often unstructured, and the environment is often extreme in terms of temperature, radiation, the vacuum of space, and the presence of dust. Important characteristics of space robotics include the observation that operators can be highly trained, but communications may be very limited due to time delays, power limitations, and even operator mobility (as in the case of an astronaut interacting from within a space suit).

Assistive robotics is a challenge area, but not because the environment is unstructured. Rather, the key attributes of this problem are the proximity and vulnerability of the human in the interaction.

Humanoid robotics is a challenge area, both in terms of engineering human-like movements and expressions, and in terms of the challenges that arise when a robot takes a human form. With such a form, social and emotional aspects of interaction become paramount.

Natural language interaction is a challenge problem, not only because it requires sophisticated speech recognition and language understanding, but also because it inevitably includes issues of mixed-initiative interaction, multi-modal interaction, and cognitive modeling.

C. Solution Themes
HRI presents a number of problems that cross application domains. These problems include requirements on autonomy, information sharing, and evaluation. Emerging from these problems are a set of solution themes that cross applications and that, when addressed, can be leveraged across several problems. In this section, we identify some of these solution themes and discuss some of the open questions associated with them.

Dynamic Autonomy, Mixed-Initiative Interaction, and Dialog. Because most interesting applications of human-robot interaction include rich information exchanges in dynamic and complex environments, it is imperative that interactions and resulting behaviors can accommodate complexity.

Telepresence and Information Fusion in Remote Interaction. Although remote control and teleoperation are the oldest forms of human-robot interaction, the problem is far from solved. In fact, with advances in robot morphology, sensor processing, and communications, it is necessary to find new ways to fuse information to provide humans an operational presence with the robot. Obstacles to achieving this include bandwidth limitations, communications delays and drop-outs, mismatches in frames of reference, communicating intent and trusting autonomy, and mismatches between expectations and behaviors.

Cognitive Modeling. Effective interactions between humans include a common ground created by common experiences and cultures. This common ground creates realistic expectations and forms the basis communications. From a robot’s perspective, supporting effective interactions also requires establishing and maintaining common ground. An emerging approach to doing this is to create cognitive models of human reasoning and behavior selection. The goal is to create rich enough models either (a) to allow the robot to identify a human’s cognitive state and adjust information exchange accordingly or (b) to allow the robot’s behavior to be generated by models that are interpretable by a human.

Team Organizations and Dynamics. Many HRI researchers are striving to develop systems that allow multiple robots and multiple humans to interact with each other. To accomplish this, it is necessary to shape team interactions and dynamics by establishing organizational structures, communications protocols, and support tools. Team organizations necessarily subsume different and dynamic roles, which implies that such efforts will need to leverage lessons from research on mixed initiative and dialog.

Interactive Learning. Because the world is complex, interactions between humans and robots are also complex. This implies that it is impossible to anticipate every conceivable problem and generate scripted responses, or anticipate every conceivable percept and generate sensor processing algorithms. Interactive learning is the process by which a robot and a human work together to incrementally improve perceptual ability, autonomy, and interaction.