Soft robotics offers adaptability and safety through compliance, enabling safe human interaction and navigation in unstructured environments. Unlike rigid robots, they can flex, absorb impacts, and adapt to various surfaces, making them well-suited for delicate tasks. We developed a bio-inspired, scalable, and biocompatible Electromagnetic Soft Actuator
(ESA). This actuator is lightweight, portable, and high-bandwidth, providing both force and deflection while overcoming the limitations of other soft actuators. However, it still faces operational challenges. A key aspect of using such actuators effectively is accurate modeling, which is essential for enabling control design, effective interactions with the environment, and task execution. However, modeling soft actuators is challenging due to their nonlinear, compliant nature [1], motivating
advanced data-driven approaches [2]. Our goal is to fabricate a modified ESA prototype to address the current operational challenges, collect experimental data, and develop an accurate data-driven model of ESA. This directly advances ARNI’s mission of integrating artificial and natural intelligence by leveraging AI-based modeling for a bio-inspired soft muscle. By characterizing the internal properties that change over time, our work will provide a natural challenge to be solved by continual learning. Separately, we are continuing to develop our previous project on soft actuator control into a pilot grant. Here, in this new project, we expect that our improved actuators will have greater capabilities, enabling us to capture a dataset useful for data-driven modeling, and eventual control, of the artificial muscle.