Natural Spatial Language Generation for Indoor Robot

This paper proposes a spatial language generation system to find short, accurate and human-like descriptions for robots to communicate with a human user about the location of an object. The research focuses on building static spatial descriptions which use reference objects and directions to describe spatial relations. The system generates a natural spatial description in three steps. In the first step, it collects the sensory information and robot state to extract an environment model. Then, it builds a grounding model that describes the location of the target object, based on landmarks in the scene. After that it will generate the natural language description by imitating a human’s talking style. A corpus of 149 spatial language commands for an indoor environment fetch task is used to train the system. An early-stage experiment was conducted and the results illustrate good potential for further development. Keywords-spatial language; language generation; robotics