ABSTRACT
Robots, like most other technological artifacts, require user interfaces for interacting with people. Interfaces for industrial robots tend to differ from interfaces for professional service robots, which in turn differ from that for personal service robots. In industrial robotics, the opportunity for human-robot interaction is limited, because industrial robots tend not to interact directly with people. Instead, their operational space is usually strictly separated from that of human workers. Interface technology in industrial robotics is largely restricted to special purpose programming languages and graphical simulation tools (Nof, 1999), which have become indispensable for configuring robotic manipulators. Some researchers have developed techniques for programming robots through demonstration (Friedrich, Munch, Dillman, Bocionek, & Sassin, 1996; Ikeuchi, Suehiro, & Kang, 1996; Mataric, 1994; Schaal, 1997). The idea here is that a human demonstrates a task (e.g., an assembly task) while being monitored by a robot. From that demonstration, the robotic device learns a strategy for performing the same tasks by itself.Most service robots will require richer interfaces. Here, I distinguish interfaces for indirect interaction with interfaces for direct interaction. For the sake of this article, I define indirect interaction to be the interaction that takes place when a person operates a robot; for example, an operator gives a command that the robot then executes. A robotic surgeon interacts indirectly with a surgical robot, as the one shown in Figure 2b, in that the robot merely amplifies the surgeon’s force. Direct interaction is different in that the robot acts on its own; the robot acts and the person responds or vice versa.A nice way to distinguish indirect from direct interaction pertains to the flow of information and control: In indirect interaction, the operator commands the robot, which communicates back to the operator information about its environment, its task, and its behavior. In direct interaction, the information flow is bidirectional: Information is communicated between the robot and people in both directions, and the robot and the person are interacting on “equal footing.” An example is the robotic caregiver in Figure 3a, which interacts with people in ways motivated by people’s interactions with nurses. In particular, it asks questions, and it can also respond to questions asked by people. As a general rule of thumb, the interaction with professional service robots is usually indirect, whereas the interaction with personal service robots tends to be more direct. There are exceptions to this rule, such as the robotic vacuum
cleaner in Figure 3c, which is a personal service robot whose interaction is entirely indirect.There exists a range of interface technologies for indirect interaction. The classical interface is the master-slave interface in which a robot exactly duplicates the same physical motion of its operator. A recent implementation of this idea is given by Robonaut (Ambrose et al., 2001), a robot developed as a telepresence device on a space station. The goal of this project is to demonstrate that a robotic system can perform repairs and inspections of space flight hardware originally designed for human servicing. Some robots are operated remotely using interfaces familiar from radio controlled cars (Casper, 2002); others possess haptic displays and control interfaces (Ruspini, Kolarov, & Khatib, 1997).In service robotics, the utility of direct interaction is much less established than that of indirect interaction. To study direct interaction, numerous research prototypes have been equipped with speech synthesizers and recognizers or sound-synthesizing devices. Some robots only generate speech but do not understand spoken language (Thrun et al., 2000); others also understand spoken language (Asoh et al., 1997; Bischoff & Graefe, 2003) or use keyboard interfaces to bypass speech recognition altogether (Torrance, 1994). Speech as output modality is easy to control and can be quite effective.Several researchers have reported excellent results for speech understanding in office robots in which speakers were instructed with regards to vocabulary the robot was able to understand (Asoh et al., 1997). Encouraging results have also been reported for a museum tour guide robot that understands spoken commands in multiple languages (Bischoff & Graefe, 2003). To my knowledge, none of these systems have been evaluated systematically with regards to the effectiveness of the speech interface. Roy, Pineau, and Thrun (2000) studied a service robot in an elderly care facility in which participants were not instructed about the robot’s vocabulary. The authors found that the speech interface was difficult to use. Only about 10% of the words used by the target group were in the robot’s vocabulary. Misunderstanding may be exacerbated if the robot’s ability to talk creates a false perception of human-level intelligence (Nourbakhsh, Rosenberg, & Thrun, 1999; Schulte et al., 1999).A number of robots carry graphical screens capable of displaying information to the user (Nourbakhsh et al., 1999; Simons et al., 2003). Researchers have used both regular and touch-sensitive displays (Nourbakhsh et al., 1999). The information on the display may be organized in menus similar to the ones found on typical information kiosks. Some researchers have used the display to project an animated face (Simons et al., 2003), such as the one shown in Figure 4a. Gesture recognition (Kahn, Swain, Prokopowicz, & Firby, 1996; Kortenkamp et al., 1996; Perzanowski, Adams, Schultz, & Marsh, 2000; Waldherr, Thrun, & Romero, 2000) has also been investigated, as has gaze tracking (Heinzmann & Zelinsky, 1998; Zelinsky & Heinzmann, 1996), as an
