An execution mechanism for natural language tasks based on auxiliary decision database

Nowadays, most of the task execution relies on the pre-set program modules, so that robots can't find solutions to new tasks autonomously. In order to improve the intelligence of robots, we propose an execution mechanism for natural language tasks based on auxiliary decision database. We first obtain resources of the semi-structured information from the Internet, and apply the Latent Dirichlet Allocation (LDA) model to classify the topics of information carefully, then select the information about the family services and store them in an auxiliary decision database. When the robot receives the service command in natural language, it can query the auxiliary decision database to obtain the steps of the task, extract the key information and generate the low-level executable instructions for each step, at last we verify the executive effect of various tasks by the Unity simulation platform. In order to assess this mechanism, we collect 1500 tasks given in the natural language from 30 users, covering four common types of service in the family environment. The results show that the robot can obtain the effective auxiliary information and action sequences for most tasks.