For the development of both hardware and software, task plannings become more and more important for robots to perform various tasks. Applying task planning to robotic system for working in real environments has difficulties. Estimating required symbols before planning is difficult because real environments are partially observable. In this paper, we proposed a method for task planning in partially observable environments with unknown objects. To construct conditional plans used in these environments, we extend the description of actions to multi-effect actions, and to deal with unknown objects, robots get new symbols generated by human interaction on demand. Additionally we show experiments of Willow Garage's PR2 executing the task in the real environment with unknown objects.

The robot system that displays detected life sound data to mobile device and learns the action corresponding to the sound正 稲葉 雅幸 (東京大)

A daily assistant robot which performs tasks in living environment should be interrupted by human at any time in order to prevent failure of task execution and improve robot's behavior. This paper provides the model of a preemptive robot action management system which is integrated with an existing distributed robot control system. This model consists of three modules which is called when a human interrupts a robot action: a preemptive task management, dynamic action state management and motion modification.