Bas Hickendorff - Enhanced human-robot interaction by reason-based behavior adaptation

afstudeercolloquium

Behavior-based robotics is successful in creating robots that are reactive enough to be used in dynamic environments. However, those systems lack the strong proactive capabilities that logic-based systems (as used in the field of multi-agent systems) have. In this research the reactive power of a behavior-based system is combined with the practical reasoning of logic-based systems to create a system for the General Purpose Service Robot (GPSR) challenge in the RoboCup@Home competitions. In the GPSR challenge, a robot can get any command that could be given in any of the other RoboCup@Home challenges. To deal with these commands, the robot should be able to handle commands consisting of several steps, commands that contain erroneous information as well as underspecified commands. In our research a system for a domestic service robot is developed that chooses the right sequence of behaviors in a flexible and robust way. The system stores scripts describing sequences of high level robot behavior and context for certain general scenarios and is able to change these scripts to make them suitable for the specific situation the robot is in. The system uses just-in-time planning to plan the details of a general plan only when it is about to execute that step of the plan. This way, the system is able to handle changing environments and failures in different components of the robot. The system is able to adapt to the users preferences and can start a dialogue about commands that are not sufficiently specific to act on, or commands that it has not heard before. The user can then explain the new command and the robot will learn a new script for that command. A preliminary version of the system was developed for the RoboCup@Home 2011 competitions in Istanbul and a full version of the system has been tested with commands from the RoboCup@Home later in the project.