Design of A Haptic Paddle for Accessible Integration of Data-Driven Methods in System Dynamics Education

Data-driven control, which embraces artificial intelligence, machine learning, and experience-based inferencing architectures, has gained significant interest for its ability to provide robust optimization in model-free, nonlinear, and time-varying paradigms. Traditional systems, such as the haptic paddle, used to communicate system dynamics principles in undergraduate curricula, have yet to be adapted to the memory and processing requirements of data-driven control. In this work, we present a modular, open-source 3D printable friction-driven haptic paddle design, building on the designs proposed by the community, using commercial components and simple microelectronic packaging, to enable robust data-driven control for integration in undergraduate education. We make use of the RP2040 microcontroller, a small light-weight logic platform capable of fast online computation and robust memory storage for onboard data-driven control. To validate our design, we first develop an experimental model of the physical dynamics that shows that our 3D printed friction drive is comparable with friction driven paddles and capstan-cable driven paddles. Further, we demonstrate the utility of our design in explicating data-driven control by presenting the development of basic machine learning and reinforcement learning architectures for online, model-free robust control in the presence of time-variable plant dynamics in a trajectory tracking task that is well suited for implementation in undergraduate and introductory graduate system dynamics and controls curricula.