Towards Intelligent Soft Grippers: Embedded Sensing and Learning-Based Multimodal Perception

Soft robotics has emerged as an innovative paradigm for tasks requiring safety, adaptability, and delicate handling. Soft grippers, in particular, have gained significant attention in agriculture, healthcare, and food handling due to their natural compliance and high adaptability. However, most current designs lack perceptual intelligence and rely solely on passive deformation rather than active sensing. This lack of integrated perception limits their autonomy and hinders their reliable use in real-world environments. This thesis addresses these challenges by developing new design and sensing strategies that integrate perception directly into soft grippers while preserving their compliance and flexibility. The first contribution details the design and evaluation of a soft gripper optimized for mushroom harvesting, demonstrating the practicality of soft robotics technologies in delicate agricultural operations. Building on this foundation, the second contribution introduces compliant sensing technologies and integration methods to enable proprioceptive feedback, establishing the basis for self-awareness and autonomous operation in soft grippers. The third contribution develops adaptive learning methods to mitigate hysteresis and drift, addressing one of the most persistent challenges in soft sensing, enabling robust long-term signal interpretation. Finally, the fourth contribution extends perception towards bioinspired exteroception, allowing soft grippers to sense and interpret object properties such as size, shape, stiffness, and roughness through embodied interaction and machine learning. Collectively, these contributions demonstrate how compliant sensing and intelligent signal processing can transform soft grippers from passive instruments into perceptually capable systems. The research advances the state of the art in soft robotic perception and outlines a pathway toward scalable, robust, and autonomous soft grippers suitable for real-world deployment.