Full Program »
Remote Laboratory For Machine Learning Training of A Soft Actuator's Control: A Case Study.
The COVID-19 pandemic continues to pose difficult challenges for engineering laboratory research. Practicing social distancing in a lab is often difficult, and as universities move academic activities online, some research labs may be affected and even shut down. This paper describes the remote laboratory implementation that the authors developed for testing a dielectric elastomer actuator (DEA) in the Soft Robotics Research Lab (SRRL) at a university campus. This system helped the authors continue developing a machine-learning controller for DEAs during the lockdown. A remote lab, as opposed to a virtual lab, does not aim toward autonomy or an immersive experience of the lab. Instead, a remote lab seeks to harness technologies already in use -- e.g., data acquisition systems and digital sensor control -- to permit students to conduct their experiments from instruments alone. After several unsuccessful collaboration solutions -- including raw secure shell (SSH) and virtual network computing (VNC) screen-sharing -- the authors selected JupyterLab as the basis for the collaboration system. This popular notebook-authoring software suite runs a Web server, and the server can be exposed over a virtual private network (VPN) to permit multiple connections. JupyterLab is an ideal platform for remote laboratory work: It provides a highly language- and platform-agnostic coding interface, as well as simple access to the host's filesystem and devices. This paper will present the remote laboratory setup, detail the remote lab's operations, and discuss results from remote experiments. The case study lays a framework for engineering educators to develop methods and practices for remote laboratory work.