Machine-Learning Driven Robot-Motion Design: Introducing a Web-Based Mechanism Design Software

This paper presents a novel machine-learning-driven web-based software, which enables the design and simulation of planar N-bar single and multi-degree-of-freedom linkage mechanisms for robotics and mechatronics applications. The software is developed using research methodologies to create a new computational framework for simultaneous type and dimensional synthesis of mechanisms for motion generation problems. The existing paradigm of selecting the type of a mechanism and then computing the dimension is shown to be inadequate in meeting the requirements of designers. Therefore, a new data-driven approach is proposed in which both the type and dimensions of a mechanism are computed directly from the user input, i.e., motion or path. While a formal assessment of the software in a classroom setting is pending, this paper outlines its broad applicability to support the learning outcomes of several mechanical engineering classes, from freshman engineering to advanced kinematics and robotics. The software has been adopted by numerous universities and organizations and was developed with funding from the National Science Foundation.