UC Berkeley

Soft robotics is an emerging field that offers adaptability and flexibility compared to rigid robots due to their inherent elasticity. Unlike rigid robots, soft robots exhibit distinct properties, kinematics and dynamics. This research, therefore, intends to introduce novel advancements in soft gripper design and control. Inspired by human skin, a dual-layer soft tactile sensor is proposed with pyramid-shaped sensing elements. It employs machine learning approaches to achieve superior sensitivity in detecting contact force, contact location, and other features. A novel design methodology for soft pneumatic actuators utilizes simplified cantilever beam approximations to develop kinematic and dynamic models. Model-based optimization design techniques are applied to determine optimal dimensional parameters, enhancing the actuator's properties such as force/torque, bendability, and controllability. To accurately capture nonlinear mechanical behavior, an alternative modeling approach using Ludwick’s Law and data-driven parameter estimation is introduced and validated through experimentation. The approach optimizes the efficiency of parameter determination within dynamical models while improving accuracy. Addressing control challenges in under-actuated systems, this research explores coordination strategies for multiple soft actuators of a soft gripper, employing stable model inversion to a multi-finger soft gripper. Simulation and experimental validation demonstrate the effectiveness of this strategy in achieving precise and coordinated motions, thereby advancing the practical applicability of soft robots. Overall, these techniques aim to enhance the capabilities of current soft grippers, paving the way for their broader deployment in real-world scenarios.