Programmable Shape morphing (PSM) devices, a pivotal subset of soft robotics, aspire to achieve programmable, controllable, and reversible transformations reminiscent of biological systems such as octopi and growing plants. They exhibit potential in realms such as augmented and virtual reality (AR/VR) devices, haptics, optical and acoustic metamaterials, and biology. However, morphing into arbitrary surfaces on demand requires a device with a sufficiently large number of actuators and an inverse control strategy.
In this talk, I will explore the integration of machine learning in achieving sophisticated control over complex shape morphing processes as part of my PhD research. Initially, I will delve into how machine learning facilitates the control of actuator arrays under complex coupling in 2D low-profile shape morphing devices. Building on this foundation, I will showcase the development of a 2D PSM device based on an array of ionic actuators. Leveraging the unique driving characteristics of ionic actuators, we have engineered a system that uses passively matrix addressing to independently control N^2 actuators with just 2N inputs. This under-actuation system significantly reduces the number of required control signals, substantially shrinking the controller size and paving the way for wearable device applications. Moving forward, I will discuss the transition from 2D to 3D PSM. Here, I will introduce how we use point cloud data to represent deformations and propose SMNet, a point cloud regression model that maps point cloud data to the inputs of actuator arrays. This approach is versatile across various types of actuator arrays and serves as a universal control framework for 3D PSM devices. Lastly, I will propose a set of performance metrics to evaluate existing studies and offer insights into future research directions. This comprehensive overview aims not only to highlight the innovative application of machine learning for dynamic shape control but also to set the stage for the next generation of PSM devices.
Jue Wang received a B.S. degree from Dalian University of Technology, Dalian, China, in 2017 and an M.S. degree from Shanghai Jiao Tong University, Shanghai, China, in 2020. He is currently a PhD candidate at the School of Mechanical Engineering, Purdue University. His research interests include machine learning algorithms, polymer actuators and sensors, computational design, and electromechanical bioreactors. He has published papers as the first author in prestigious journals such as Science Advances, IEEE Transactions on Robotics, Advanced Intelligence Systems, and IEEE Robotics and Automation Letters, among others. He is set to complete his PhD degree this summer and is currently seeking postdoctoral positions.