ABSTRACT
The development of humanoid robots such as Honda’s ASIMO (Hirai, Hirose, Haikawa, & Takenaka, 1998) and interactive robots such as Sony’s AIBO® (Fujita, 2001) and Kismet (Breazeal & Scassellati, 1999) has spawned a new area of research known as interactive robotics. These are not robots performing simple iterative tasks in factories or using specific tools in professional services such as surgical or military tasks (Thrun, 2004). Rather, this new wave of research is exploring the potential for partner robots to interact with people in daily life. Our research explores some fundamental problems in this new field.Several researchers and companies have endeavored to realize robots as partners for people, and the concept of a partner robot is rapidly emerging. Typically equipped with an anthropomorphic body and various sensors used to interact with people naturally, the partner robot acts as a peer in everyday life. A humanoid robot, for example, guides office visitors by speech and with a hand-gesture recognition mechanism (Sakagami et al., 2002). For the home environment, NEC Corporation (2002) developed a prototype of a personal robot that recognizes individuals’ faces, entertains family members with its limited speech ability, and performs as an interface to television and e-mail. Partner robots have also appeared in therapeutic applications. For example, Dautenhahn and Werry (2002) are applying robots to autism therapy. As these examples show, partner robots are beginning to participate in human society by performing a variety of tasks and functions.Eliza was the first computer agent that established a relationship as a partner (Weizenbaum, 1966). People tried to interact with Eliza without necessar
ily having a specific task or request in mind. They sometimes made brief small talk and at other times engaged deeply in conversation. As Reeves and Nass (1996) discovered, humans unconsciously behave toward such a computer as if it were human. In recent robotics research, several pioneering studies have suggested that humans also can establish relationships with pet robots. Many people actively interact with animal-like pet robots. For example, people have adapted to the limited interactive ability of the robot dog, AIBO (Friedman, Kahn, & Hagman, 2003; Fujita, 2001). Furthermore, pet robots have been used successfully in therapy for the elderly, with some positive effects of their usage confirmed in long-term trials (Wada, Shibata, Saito, & Tanie, 2002). 1.2. Social Relationships Over Time
Recognizing the other person’s identity, discovering similarities, and finding common ground are key issues in cementing social relationships. As Isaacs and Clark (1987) proposed, when people first meet, they gradually establish common ground through conversation. Empirical studies have shown that interlocutors adapt their speech to each other’s attitudes and experience, weighing each other’s perspectives when listening and making themselves understood (Fussell & Krauss, 1992). In forming satisfying and stable intimate relationships, they may even find similarities in their partner that do not exist in reality and tend to assume that their partner is a mirror of themselves (Murray, Holmes, Bellavia, Griffin, & Dolderman, 2002).This evidence shows the importance of finding common ground in establishing relationships. However, relationships among people evolve over time (Hinde, 1988), and we believe people’s attitude toward technological artifacts and their relationship with them also evolves over time. Little previous research has focused on long-term relations between individuals and computer systems in general or partner robots in particular. Short-term and long-term analyses must be carried out to evaluate partner robots. With respect to short-term experiments, many evaluation methods and systems have been proposed within the field of human-computer interaction and robotics. For instance, Quek et al. (2002) developed a gesture recognition-based system to analyze multimodal discourse. In robotics, Nakata, Sato, and Mori (1998) analyzed the effects of expressing emotions and intention. We have also performed several similar experiments, such as examining the effects of behavior pattern on impressions (Kanda, Ishiguro, & Ishida, 2001; Kanda, Ishiguro, Ono, Imai, & Nakatsu, 2002). However, in short-term human-robot interaction, we can only observe first impressions and the initial process of establishing relationships.Some previous research has stressed the importance of long-term studies. Fish, Kraut, Root, and Rice (1992) evaluated a videoconferencing system and
analyzed the transition of system use during 1 month of experimentation. Petersen, Madsen, and Kjaer (2002) reported on the process of gaining experience with a new television system. These studies showed that the relation between human and agent is likely to change over time, just as interhuman relationships do. Therefore, it is vital to observe relationships between individuals and partner robots in an environment where long-term interaction is possible. The result of immersing a robot in an environment that demands ongoing participation is likely to be entirely different from that of exhibiting the robot in a public place like a museum, where the people who interact with it are transient. 1.3. Technologies for Creating Human-Robot Relationships
As previous research on interpersonal communication indicates, it is vital that two parties recognize each other for their relationship to develop. We cannot imagine having human partners or peers who cannot identify us. It is because we are able to identify individuals that we can develop a unique relationship with each of them (Cowley & MacDorman, 1995; Hinde, 1988). Although person identification (ID) is an essential requirement for a partner robot, current visual and auditory sensing technologies cannot reliably support it. Therefore, an unfortunate consequence is that a robot may behave the same with everyone.Given only visual and auditory sensors, it is difficult to implement a person ID mechanism in robots that works in complex social settings. Many people may be talking at once, lighting conditions may vary, and the shapes and colors of the objects in the environment may be too complex for current computer vision technologies to function. In addition, the method of ID must be robust. Misidentification can ruin a relationship. For example, a person may be hurt or offended if the robot were to call the person by somebody else’s name. To make matters worse, partner robots that work in a public place need to be able to distinguish between hundreds of people and to identify nearby individuals simultaneously. For instance, consider a situation involving people and robots working together in an office building, school, or hospital.Besides their ability to identify and recognize others, robots should have sufficient interaction ability. In particular, human interaction largely depends on language communication. Whereas speaking is not so difficult for the partner robot, listening and recognizing human utterances is one of the most difficult challenges in human-robot interaction. Although some of the computer interfaces successfully employ speech input via microphone, it is far more difficult for the robots to recognize human utterances, because the robots suffer from noise from surrounding humans (background talk) and the robot body (motor noise). Little research has reported the solutions to this serious problem. We cannot expect ideal language perception ability like humans. How
