Fellow: Naren Vasantakumaar
Supervisor:
Co-Supervisors:
How can uncertain conditions be modelled for random simulations from the implied probability distributions? How can we automatically transform a distribution of dtKR&R simulations into joint probability distributions over hierarchically structured hybrid symbolic/continuous automata?
We will transfer modern Knowledge Representation and Reasoning (KR&R) techniques for introspective reasoning using probabilistic distributions in the field of cognitive robotics to the problem domain of self-explanation. These KR&R techniques, called digital twin Knowledge Representation & Reasoning (dtKR&R), KR&R (dtKR&R), enable agents like robots to accomplish serve underdetermined task requests such as ‘‘set the table’’ and ‘‘bring me something to drink’’, to perform an action without causing unwanted side effects, and understand what they are doing, why, and how. They represent agents and their environments, that is a subclass of CPSs, as virtual reality scene graphs, which facilitate the photo-realistic rendering and physical simulation. dtKR&R also links substructures of the scene graph to probabilistic distributions to find the relation between them, and perform causal inference for explainability.