We consider here the problem of connecting natural language to the physical world for robotic object manipulation. This problem needs to be solved in robotic reasoning systems so that the robot can act in the real world. In this paper, we propose an architecture that combines grounding and planning to enable robots to solve such a problem. The grounding system of the architecture grounds the meaning of a natural language sentence in physical environment perceived by the robot’s sensors and generates a knowledge base of the physical environment. Then the planning system utilizes the knowledge base to infer a plan for object manipulation, which can be effectively generated by an Answer Set Programming (ASP) planner. We evaluate the overall architecture on several datasets and a task of RoboCup2014@home (http://www.robocup2014. org/). The results show that the new architecture outperformed some other systems, and yielded acceptable performance in a real-world scenario.
CITATION STYLE
Lu, D., & Chen, X. (2015). Towards an architecture combining grounding and planning for human-robot interaction. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9513, pp. 214–225). Springer Verlag. https://doi.org/10.1007/978-3-319-29339-4_18
Mendeley helps you to discover research relevant for your work.