Attentional Mechanism for Sensory-motor Coordination in Behavior-based Robotic Systems

This thesis focuses on the problem of efficiently allocating resources for enhancing the performance of an autonomous robotic agent. Such an agent is expected to operate in complex dynamic environments by continuously monitoring its internal states and the external events. These requirements give raise to countless problems that have populated research in the autonomous robotics community in the last two decades. Among these issues, one of the most relevant is to coordinate different low and high-level behaviors, giving them, from time to time, different priority values both for resource allocation and for action selection processes. The main problem in achieving this requirement is that the number and complexity of the stimuli received by each behavior may be quite high and also the effects on the emerging activity may be very hard to foresee. It is clear that it is not possible for the robotic control system to process all the incoming information, especially for real-time applications. Thus, it becomes necessary to build mechanisms able to guide this sensory input selection process and to choose the best action to perform, assuring an efficient use of the robot limited sensorial and cognitive resources. For this purpose, attentional mechanisms, balancing sensors elaboration and actions execution, can be very useful since they play two main roles: they focus the attention on salient regions of the space and they distribute resources and activities in time. As a result of the application of these mechanisms within the robotic control system for sensory-motor coordination, the robot behavior is improved: the robot becomes able to react faster to task-related or safety-critical stimuli and to opportunely split resources among concurrent behaviors. Attentional mechanisms applied to autonomous robotic systems have already been proposed, but mainly for vision-based robotics; conversely, the contribution introduced by the present thesis is the use of an artificial attentional mechanism suitable both for optimizing the use of resources and for execution monitoring and control.