Marco Agumentativo Estructurado Generalizado
Definición 22 (Soporte entre Átomos Argumentales) Sea
Robots were once programmed to perform repetitive actions in an open-loop manner in industrial settings that had been specially designed for them, and only certain people were allowed near robots in order to reprogram or repair them. However, because robots can now operate autonomously or be tele-operated in a variety of settings and are often asked to work with and interact alongside people, researchers have begun to study how humans and robots can successfully collaborate with each other and to design paradigms and protocols for doing so.
Robots and humans interacting as equals
Drawing inspiration from how groups of people work together to solve problems, some human-robot interaction researchers have studied how groups of people and robots collaborate together for mutual benefit. Hinds, Roberts, and Jones studied the effects of different robot appearances and different robot status roles on the task-solving capabilities of different human-robot collaborative pairs. Ultimately, they found that people collaborated better with human-like robots when the people
had to delegate responsibilities to them or ask them to perform tasks too demand- ing for people, while more machine-like robots and people collaborated better in situations where the robot had a higher chance of being unreliable or where per- sonal responsibility was emphasized [Hinds et al., 2004]. Drury, Scholtz, and Yanco outlined an awareness framework for describing different human-robot collaboration scenarios and were able to re-evaluate different failures in a collaborative human- robot search-and-rescue competition in terms of various deficiencies in human-robot awareness [Drury et al., 2003]. Later, Yanco and Drury outlined another framework that categorized all forms of human-robot collaboration according to how coopera- tively the humans and robots behaved in temporal as well as spatial terms; how the humans and robots communicated their collaborative efforts to each other; the orga- nization of the robots themselves; and the nature of the human-robot collaborative task itself [Yanco and Drury, 2004]. Sidner, Lee, and Lesh studied how robots could use conversational gestures and gaze patterns to better engage and sustain people in collaborative, socially assistive interactions. By focusing on turn-taking, interpret- ing human actions, and shared-goal decision-making, they adapted human-human collaborative conversation techniques in order to use in human-robot collaborative conversation [Sidner et al., 2003].
Human as manager of robots
Laengle, Hoeniger, and Zhu examined how robots and humans could work together in groups and determined that such groups had four requirements: intelligible com- munication among all parties, proper interpretation of what was communicated, co- ordination of activities, cooperation among the agents when teamwork is required, and safety precautions implemented on the robots for the benefit of the humans. In these groups, the humans would be coordinators and managers for the robot workers, and such principles seemed to have promising results when implemented on the KAMRO robot and its multi-agent architecture, KAMARA [Laengle et al.,
1997]. Fong, Thorpe, and Baur studied how a properly designed human-robot com- munication protocol could lead to an easier-to-use collaborative robot teleoperation system (using the definitions described earlier in the thesis, such a system should re- ally be called a cooperative system). Specifically, by having humans cooperate with multiple independently-controlled robots, both were able to accomplish more than the instances where the humans had to manually control every aspect of the robots [Fong et al., 2001]. In a follow-up user study on the same system, the researchers found the the roles and capabilites of the human controller and robots should have been more clearly defined. Additionally, the study suggests that control strategies should be developed that would not require a human to control the general move- ment of any single robot, such as global (large-scale swarm) control. Such a strategy should also allow a human to easily resume control of a single robot to determine the cause of robot problems via a dialogue system while increasing the autonomy granted to other robots [Fong et al., 2003b].
Leonardo
Breazeal, Hoffman, and Lockerd developed a humanoid robot Leonardo which was capable of being taught in naturalistic ways by people to execute simple tasks and then performing them collaboratively with another person [Breazeal et al., 2004]. Leonardo communicated with partners via social gestures and facial expressions, and its sensing as well as learning systems were motivated by the idea of joint in- tention, or multiple agents continually coordinating their actions and intentions in order to collectively achieve a mutual goal (e.g. executing a coordinated attack, maneuvering of a heavy object) [Levesque et al., 1990]. Lockerd and Breazeal found that Leonardo’s speed in successfully learning a simple task (collaboratively pressing buttons with another participant) using its technique of socially-guided learning was superior than its expected performance if it had used Q-learning in a number of dif- ferent configurations [Lockerd and Breazeal, 2004]. Later, Breazeal, Kidd, Thomaz
et al analyzed self-report questionnaires and coded video footage from people teach- ing and collaborating with Leonardo under two different conditions: in one case, the robot would continually express its internal state both implicitly via gaze as well as other non-verbal behaviours and explicitly using expressive gestures and other social cues, while in another, the robot would only express its internal states using expressive gestures when it was asked to do so. According to the questionnaires, the researchers determined that participants understood the robot’s abilities and states at any given time as well as understood what the robot was “thinking” better when it communicated implicitly as well as explicitly. Furthermore, the coded video footage showed that the humans took less time interacting with the robot overall, found errors in its behavior faster, and corrected the errors better when the robot communicated both explicitly and implicitly [Breazeal et al., 2005].