Autonomous behavior control subsystem for robotic computer

Successful development of a robotic computer as a mediator in smart environments requires providing a certain level of behavior autonomy to the robot and a capability to adapt its behavior in long-term interaction with the users. We attempt to identify core autonomy-related functionalities and describe the design and implementation of an autonomous behavior control subsystem that provides them. The choice of dual-level behavior model with both rules and connectionist reinforcement learning is motivated by the desire to provide convenient specification of initial behaviors as well as efficient behavior adaptation capability. We present the results of two test scenarios illustrating basic use of the newly provided functionality.