PaGDeSA: An Automated Generative Design of Multimodal Soft Pneumatic Actuators

The recent advancements in machine learning techniques have fostered a growing emphasis on data-driven design. In soft robotics, this often involves optimizing for specific tasks rather than learning the design space. Motivated by this, the present study proposes a Performance-augmented Generative Design of Soft Actuators (PaGDeSA), an automated data-driven framework that explores the design space to generate new soft actuator designs. As a first step, a synthetic dataset of soft Pneumatic Network (Pneu-net) actuators has been created using data augmentation to train the generative model. Following on, the parametric design of two different unimodal (namely, pure bending and twisting type) actuators is used to train the Gaussian mixture model that generates novel actuator designs to perform in-plane and out-of-plane deformations. A distance-based metric is used to describe the novelty and diversity of the generated designs. Later, the feasibility of these designs is ensured using an automated finite element simulation. Further, the feasible generated design is passed on for digital fabrication to assure the manufacturability of the actuator design. In addition, a comparison of the bending angle (quality metric) between the simulated and real design strengthens the proposed approach. The framework could accelerate the design of new soft robots by enabling the selection of appropriate actuators from the diverse set of generated multimodal actuators.