| Robot control software presents certain unique challenges: it is often distributed among many processors, and frequently consists of many independent interacting modules. Also, because the robot often interacts with dynamic or unpredictable environments, the robot may not always achieve its goals. Furthermore, development of robots is shifting towards systems that are more intelligent, autonomous, and user-friendly, particularly in the case of humanoid robots.; To this end, a framework for human-humanoid interaction (HHI) in a multi-agent system is being developed. This dissertation describes the design and implementation of part of that framework, specifically a high-level virtual agent, the Self Agent. The Self Agent acts as part of the third, or deliberative, layer of the humanoid's control software, providing support for deictic interaction, a model of the humanoid's system status, and an algorithm for arbitrating between the intentions of multiple humans. This model includes status information gathered from other agents in the system via low-level and task-level monitoring processes. Experiments show that the monitoring processes can detect certain failures in the agent system. Further experiments demonstrate the humanoid's ability to arbitrate among the conflicting intentions of multiple humans. |
