The human agent: a model for human-robot interaction

摘要

Many people are familiar with the version of robots, namely humanoids and androids, depicted on film or in print. As the realized versions of robots are becoming more sophisticated, so are people's expectations of robot performance and capabilities. However, these expectations still vary greatly and can affect how people interact with robotic systems. This dissertation presents results of work aiming to raise the robot's interaction abilities. At Vanderbilt University, we have developed a multi-agent based cognitive robot architecture. The Human Agent is the element of this architecture that represents the people in the robot's environment. It includes information about their physical and cognitive states based on observations and conversations. The Human Agent is defined as a virtual agent that serves as an internal active representation of people in the robot's environment. As a representation, it is able to detect, represent, and monitor people. The description active is used, much as in describing active perception vision systems [Bajcsy 1987], to indicate that the system can take action to make its representation richer. It takes action by determining appropriate social interaction behaviors. The design of the Human Agent is presented. It performs the two main roles of monitoring and interacting with people. Monitoring is based on several inputs, including detection of motion and sounds, the recognition of faces and voices and the interpretation of speech. For interaction, the Human Agent is equipped with a set of rules for social interaction and for selecting socially appropriate behaviors. Results of realizing this design on the ISAC humanoid and deploying it in an interaction environment are presented and discussed. An assessment of the robot's activity and its social appropriateness is also presented. As an integrated entity in the robot architecture, the Human Agent allows the robot to become aware of people, interact with them appropriately, and accomplish tasks.