Generative model-based control strategies for soft robotics arms

This research work delves into the domains of soft robotics and generative artificial intelligence, proposing innovative methodologies to tackle the challenges within these fields. Soft robotics, characterized by manipulators crafted from soft materials, offers promising prospects for safe human interaction. Simultaneously, the recent advancement of Generative AI has revolutionized content creation and manipulation. Addressing the complexities of controlling soft manipulators, this research investigates the possibility of using generative AI to control soft robotics arm by introducing two pivotal contributions. Firstly, the KineFormer, a generative model based on transformer architecture aims to model the inverse kinematics of soft manipulators. This model displays superiority over the baseline's models, ensuring consistent limb shapes during control. Secondly, a system leveraging reinforcement learning with DynaFormer, a transformer-based forward dynamics model, is proposed. This system showcases its effectiveness in controlling soft robotics arms in dynamic settings. Overall, this research pioneer's novel approaches the intersection of soft robotics and generative AI, offering promising avenues for shaping the future of adaptable and versatile robotic systems across diverse practical domains.