Knowledge Representation & Reasoning

4. Conclusion

Knowledge representation and reasoning provides both, huge opportunities for the investigation of intelligent robots as well as huge challenges for KR&R research. The opportunities are well stated by Pratt, 2015: "Robots are already making large strides in their abilities, but as the generalizable knowledge representation problem is  addressed, the growth of robot capabilities will begin in earnest, and it will likely be explosive.'' The challenges become obvious if we look at research in the field of action science. The brain has orchestrated very diverse, sophisticated, and powerful reasoning techniques to achieve the cognitive capability to accomplish complex goal-directed object manipulation capabilities. The representation and reasoning capabilities include simulation-based (Hesslow, 2002), probabilistic (Griffiths, 2008), imagistic (Kosslyn, 2005) reasoning, and prospection (Vernon, 2015). In addition, other information processing mechanisms including situation models that enable flexible, context-specific, goal-directed behavior as well as memory mechanisms (Tulving, 1972) are not understood well enough to build informative models of their operation. Other key research questions that deserve intensive research activities are how to ground reasoning techniques into the perception and action mechanisms of robot agents and how to combine symbolic representations with task-directed representations automatically generated through data intensive machine learning techniques.