Knowledge Representation and Reasoning in Robotics
Papers from the 2014 AAAI Spring Symposium
Mohan Sridharan Program Chair
Robots and agents deployed in homes, offices and other complex domains are faced with the formidable challenge of representing, revising and reasoning with incomplete domain knowledge acquired from sensor inputs and human feedback. Although many algorithms have been developed for qualitatively or quantitatively representing and reasoning with knowledge, the research community is fragmented, with separate vocabularies that are increasingly making it difficult for these researchers to communicate with each other. For instance, the rich body of research in knowledge representation using logical reasoning paradigms provides appealing commonsense reasoning capabilities, but does not support probabilistic modeling of the considerable uncertainty in sensing and acting on robots. In parallel, robotics researchers are developing sophisticated probabilistic algorithms that elegantly model the uncertainty in sensing and acting on robots, but it is difficult to use such algorithms to represent and reason with commonsense knowledge. Furthermore, algorithms developed to combine logical and probabilistic reasoning do not provide the desired expressiveness for commonsense reasoning and/or do not fully support the uncertainty modeling capabilities required in robotics.