To be reliable and efficient, service robots must flexibly interleave their tasks, exploit opportunities, quickly plan their courses of action, and, if necessary, revise their intended activities. The limited computational resources available to robots prevent them from recomputing their best possible courses of action whenever situations change. The consumption of computational resources can often be drastically reduced if the robots' controllers execute plan-based control , that is, if they explicitly manage the robots' beliefs and current goals and revise their action plans accordingly.

In recent years, there have been a number of long-term real-world demonstrations of plan-based control, which have impressively shown the potential impact of this technology for future applications of autonomous service robots. A plan-based robot controller, called the Remote Agent , has autonomously controlled the performance of a scientific experiment in space in NASA's Deep Space program. In the Martha project, fleets of robots have been effectively controlled and coordinated. Xavier , an autonomous mobile robot with a plan-based controller, has navigated through an office environment for more than a year, allowing people to issue navigation commands and monitor their execution via the Internet. In 1998, Minerva acted for thirteen days as a museum tour guide in the Smithsonian Museum, and led several thousand people through the exhibition.

Researchers have now begun to tackle new application areas, such as the plan-based control of robot soccer teams, and long-term application challenges, such as the robotic assistance of elderly people and the plan-based control of robotic rescue teams after earthquakes. More uses of plans in robot control than goal-based organization of a course of action are coming into focus, such as

• robot/human and robot/robot interaction,
• high-level task-oriented command or tele-manipulation interface languages,
• response to unforeseen changes in the robot environment or its mission goals,
• defining an abstract description layer for robot controller design,
• chunking or learning complex behavior structures.

This Dagstuhl Seminar brings together a team of leading researchers who are investigating different aspects of plan-based control. They will discuss issues in plan-based control and work towards a common framework for plan-based control of robotic agents.

Scope

The Dagstuhl Seminar will focus on the following key issues in plan-based control of robotic agents:

• Plan execution and monitoring. What is the right representation and expressivity of plans for autonomous robot control? How can a plan layer support flexible and reliable execution? What kind of plan management operations other than plan formation should a plan layer provide? How can the plan representation be grounded in the sensor and actuator capabilities of the robot?
• Execution-time plan management. How can we make plan management operations feasible? Should plan management be incorporated into a plan layer of a hybrid robot control architecture? Are there other means for integrating plan management into the overall control?
• Plan-based control and learning. How can a robot automatically chunk its continuous behavior and learn complex behavior structures? How can it automatically adapt to specific environments and tasks? How can it learn or acquire the plan schemata and planning knowledge needed for execution time plan management?
• Formal models of plan-based control. What are realistic formal models of plan-based control? Could or should they include plan revision and execution-time plan management? Can formal languages be used to define an abstract description layer for robot controller design? Do formal models help clarify robot/human and robot/robot interaction, high-level task-oriented command, and tele-manipulation interface languages?
• Robot mission planning applications. What applications of long-term high-level plan-based robot control exist? What are emerging applications? Do they comprise scheduling functionality?
• Challenge problems and benchmarks. What are challenge applications for plan-based control of robotic agents? Are there empirical results on the use of plan-based versus behavior-based robot controllers? Should we try to collect benchmarks for plan-based robot control?

More Information is available in http://www.ais.fraunhofer.de/ARC/Dagstuhl/dagstuhl01431Details.html