Development of intentions recognition system to control mobile robot with input disturbance

Abstract

This project explores on the collaboration between human and robot. It aims to understand human’s intention with the tolerance to uncertain or unsteady input due to user’s physical impairment. Existing Intention Recognition System (IRS) is not able to recognize the unsteady input. Hence Intention Recognition System with filtering ability (IRS-F) is developed. The Bayesian theorem and Dynamic Bayesian Network are applied for the inferring procedure with the aid of features recognition module. For implementation, the machine learning database is rearranged with more precise velocity states to identify more delicate changes of robot. All the possible actions of robot are classified into appropriate intentions. To create tolerance, the concept of fuzzy logic is adapted into machine learning database by creating the overlap region. The experiments were conducted in Robotic Research Centre (RRC) to test the performance of categorized motions which are combination of translation and rotational motions. Vibration input of joystick is used to simulate the input disturbance to test the ability of tolerance. The results showed that intention recognition system possessed the ability to filter the input without eliminating any information. As a conclusion, the objective of project is achieved by restructuring the machine learning database and integrating the overlap region between intentions. However, further improvement can be added to aid in the robustness and the capabilities of the systems.