Abstract

Soft robots have unique properties such as inherent compliance, safety, and adaptability, which make them attractive for applications in unstructured and human-centric environments. However, their nonlinear mechanical properties and geometric complexity pose significant challenges in modelling, design, and simulation. In addition, the unique qualities of cable-driven grippers which mimic muscle tendons, provide high force density and greater control, bring their own challenges. This thesis addresses three critical interrelated challenges which limit the scalability and reliability of tendon-driven soft robotic grippers: (1) the lack of accurate cable friction modelling for elastic surface contacts, (2) the cost and inefficiency of the design process for task-specific grippers, and (3) the absence of real-time simulation tools that accommodate hyperelastic and multi-material behaviour.
First, a novel friction model is proposed that captures the asperity behaviour between cables and elastic surfaces. Unlike existing friction models developed for rigid systems, this formulation accounts for deformation-dependent force transmission and can be calibrated with as few as nine data points. When validated experimentally, the model reduced tip prediction error from 16.1% (baseline) to 2.8% for a soft robotic finger with three joints.
Second, a grasp-based product classification framework is introduced. The frame-work maps food items to a small set of human-inspired grasp types. This classification supports a modular and reconfigurable gripper design strategy that balances versatility with task-specific performance. A streamlined design pipeline integrates human demonstration data, kinematic modelling, stiffness and cable placement optimization to rapidly generate custom gripper configurations. The resulting modular gripper was validated across 15 diverse food items, achieving trajectory tracking accuracy exceeding 97%, with a reconfiguration time under five minutes and full fabrication cycles under 24 hours.
Third, to support simulation-informed design and control, a novel geometry-based simulation framework is developed to efficiently model nonlinear, hyperelastic deformations. By embedding strain energy-dependent stiffness into element-wise parameters, the approach dynamically captures material behaviour without updating the global stiffness matrix, thereby maintaining the computational speed advantage geometry-based solvers have. This method enables rapid simulation of complex geometries with arbitrary amounts of materials. Thus, addressing current limitations which restrict existing geometry-based methods to two linear materials.
Collectively, the work presented in this thesis contributes new theoretical models, computational methods, and experimental frameworks that enable faster, more reliable, and more adaptable design of soft robotic grippers. These contributions address key bottlenecks in friction characterization, design scalability, and material simulation, and provide a pathway toward broader industrial adoption of soft robotics.