Augmenting Compliance With Motion Generation Through Imitation Learning Using Drop-Stitch Reinforced Inflatable Robot Arm With Rigid Joints

Abstract

Safe physical human-robot collaboration is possible with soft robots due to their inherent compliance and low inertia. Soft bodies inherently possess passive compliance, providing adaptability in collaborative tasks because of their deformations; however, the same features add complexity to modeling and dynamic control. We focus on motion generation for a 3 degrees of freedom (3DOF) inflatable robot arm, which consists of soft inflatable body links and rigid joints. This research explores the limitations of relying solely on soft robot compliance for completing contact-based tasks. Our goal is to generate adaptive motion for contact-based tasks by exploiting the compliance of soft links. We compare contact-based tasks involving an inflatable robot with and without a learning model. Improved performance is achieved when soft robot compliance is augmented with imitation learning. The combination of soft robot compliance and the adaptability of the machine learning model demonstrates the potential for collaborative robots to safely interact with humans and their surroundings.