Symbolic Control System for a Mobile Robotic Platform Based on SOAR Cognitive Architecture

Abstract

Problem statement: It is important to be able to share various types of information between robot modules, including high-level data received from the user and low-level signals from the sensors. This paper describes an implementation of such a system that uses semantic networks as a generic representation providing interpretability of instructions and flexibility in parameterization. Purpose of research: development of a robotic control system based on standard libraries for logical processing that can execute complex commands including high level instructions derived from human speech that involve objects, spatial relations and clarifying information. Results: The proposed semiotic control system consists of a database that stores knowledge about the robot's environment and its behavior using the RDF data model, inference tools, a pipeline of filter modules and the corresponding interfaces. SOAR provides a standard representation of facts and rules. The external interface of the control system based on the semiotic model transforms natural language commands into semantic networks. It can be easily expanded to support other input signals as eyetracking or encephalograph due to the ability to include clarifying information as extra nodes of the intermediate semantic network. The developed system can execute both direct control commands such as movement in a specified direction, and more complex procedures like moving to the named object with a heuristic choice between alternatives and parameterization of the trajectory. Practical significance: Using natural language voice commands, intonation, and looking at the target is essential for effective operator interfaces with mobile platforms and has applications in many areas including assistive and service robots. Using standard implementations and frameworks for logical processing makes the system more reliable, efficient and easier to understand.