Design and Control of Compliant Robotic Systems for Hand Rehabilitation and Tendon-Driven Continuum Robots

This thesis investigates the design and control of compliant robotic systems for human-centred healthcare, with a dual focus on wearable hand rehabilitation devices and tendon-driven continuum robots (TDCRs) for minimally invasive interventions. The aim is to develop robotic systems that remain mechanically feasible under realistic practical constraints while maintaining compatibility with human anatomy. The first part focuses on hand rehabilitation. A finger exoskeleton architecture is proposed for the metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joints based on Remote Centre of Motion (RCM) mechanisms. The linkage generates virtual centres aligned with anatomical joint axes. Most of the mechanism remains within the hand profile, helping keep the device compact and wearable. To provide adjustable resistance, a pneumatic variable-stiffness layer is developed using 3D-printed inflatable chambers integrated into a glove. The chamber design was gradually improved through prototyping, from early Flexojoint concepts to a compact pillow-shaped geometry. Experiments were conducted to quantify the influence of bending angle on internal pressure, reaction force, and resistive torque across a 0◦–45◦ range under various initial inflation pressures in a sealed pneumatic circuit. These experiments show pressure-tunable torque–angle profiles that can support different rehabilitation stages. The second part focuses on the design, planning, and control of tendon-driven continuum robots under practical constraints. First, a model-based framework is developed to analyse static feasibility using a discretized PCC representation and a kinetostatic formulation that accounts for tendon forces, gravity, friction, backbone elasticity, and external loads. This framework is subsequently used to analyse the feasible static workspace and to optimise tendon-routing parameters so as to maximise workspace feasibility while minimising mechanical effort. Based on these results, two planning strategies are introduced: a hybrid inverse-kinematics scheme combining supervised learning with constrained refinement, and an offline capability-map framework that stores feasible voxel-orientation targets together with dexterity information for runtime selection. Finally, prescribed-time safe and fault-tolerant control strategies are developed, including observer-based estimation and learning-assisted compensation under uncertainties, delays, actuator faults, and input saturation.