Rule-based supervisor and checker of deep learning perception modules in cognitive robotics

This paper proposes the adoption of rule-based supervisor with logical agents as implementation of a reasoning layer embedded in cognitive robotic controllers where the world model representation is built from the output of a layer of deep learning perception modules. In this way the logical augmented cognitive robot could self-check in situations were safety and security issues have an impact in applications like user monitoring and robot navigation in shared spaces with humans. In previous related work we have suitably augmented the implementation of the DALI language, and we are presently developing promising sample applications in real robotic settings on which we shortly report here.