A distributed architecture for interactive robots based on a knowledge software platform A Distributed Architecture for Interactive Robots Based on a Knowledge Software Platform

博士論文

Robots have taken more and more important roles in human life. That is, they are not only working in factories and hazardous environment but also being introduced in applications like welfare services and amusement agents, where interaction with human is crucial. In the future, we envision interactive robots assisting humans especially the elderly and the disabled in daily life. Such robots coexist and interact with humans in a human-friendly manner. <br />　To develop such robots we need to address two main problems of how to integrate various robotic components, and how to control robot behaviors so that it interacts with humans reasonably. The first problem involves architectural design of the system in the component level, i.e., how robotic components can be combined into a single system. The second problem involves designing the brain part of the system, which manages robot actions and interactions with human. There are several approaches in designing robot architectures to accommodate these problems, e.g., behavior-based, plan-based, and knowledge-based approaches. In our research group, we have been focusing on the knowledge-platform-based approach using the frame model, and have developed a software platform for robot system development called the Software Platform for Agents and Knowledge(SPAK), in which a general purpose software platform integrating various robot devices and applications is placed at the center of the system. <br />　In this work, we extended SPAK to overcome the problems of representing changing knowledge and limitations in managing human-robot interactions found when deploying SPAK in a robot system. New extensions to the conventional frame model, namely, time-based layer, periodical task evaluator, and priority support for frame-related actions, are proposed and added to SPAK. With the time-based layer, the knowledge manager keeps track of changes of the knowledge contents and provides methods to access this history data. One can, for example, query a frame's age, old values of a slot at an absolute or relative point of time, and set the frame's condition based on this history data. The periodical task evaluator regularly checks the validity of the knowledge contents, fixes it if needed, and executes actions. Designers can specify actions to be done periodically in an event-driven slot, whose contents will be executed by the evaluator. Priority support for frame-related actions allows the knowledge designer a fine-grain control of the execution order of frame-related actions. Each frame can be assigned with a priority value in a system slot. SPAK engine will respect these values when conducting actions like frame instantiation, updating, deactivation, and evaluations. These new extensions can be used via new special system and event-driven slots, and slot flags added to SPAK. <br />　To very the utility of SPAK and to illustrate how a robot system can be built based on this architecture, a demonstration system on a humanoid robot was developed, with a sample dialogue management application. The knowledge model for the dialogue manager is shown. The current knowledge-based dialogue manager can handle basic human commands, state-based and form-filling dialogues. A multi-agent technique is employed to connect various robotic hardware and software components designed as agents together on the network. The prototype system was set to make dialogue interactions with a human in three scenarios: basic learning, greeting and future welfare robot. <br />　This dissertation makes three contributions to the fields of knowledge engineering, dialogue systems, and robotics. <br />　First, the thesis proposes a design concept for the knowledge manager, the brain part of the robot. We based our design on the frame knowledge model because of its simplicity and naturalness, and introduced dynamic extensions to the conventional frame model to support robot behaviors control. Various robotic applications can run and share knowledge among each other on the platform. <br />　Second, the thesis proposes a novel design of a knowledge-based dialogue manager for robot as an application on the platform. The design and internal mechanism are natural, easy for developers to understand. With tight integration with the knowledge base, the dialogue manager can easily make use of the knowledge and facilities provided. <br />　Third the thesis contributes a knowledge-based robot architecture with an implementation as a robot system that interacts with humans in the laboratory environment. The system components include SPAK, the designed dialogue manager, various software and devices, and a Robovie humanoid robot. With SPAK we show that it is simple and intuitive to develop a multi-modal interactive robot system. <br />　A SPAK-based robot system has an important feature of frames and agents being integrated seamlessly in a distributed environment. A prototype system with a sample dialogue management application has demonstrated interesting functions for future symbiotic robots. Ultimately, this dissertation demonstrates the concept of employing a knowledge platform as the base layer of a robot system. The platform works as glue connecting various robotic devices and applications together. We believe that this is a major step forward towards achieving intelligent future robots. <br />

application/pdf

総研大甲第903号