Customizable and Reconfigurable Continuum Robots via Stiffness Encoding

Abstract

Biological soft manipulators, such as octopus arms, achieve dexterous manipulation by coordinating muscle bundles that move and stiffen the limb, adapting motion primitives to the task. Inspired by this, soft and continuum robots use continuously deformable structures to navigate constrained spaces while minimizing the risk of damage to their surroundings because of their compliance. This dissertation advances the design, modeling, and actuation of such robots by considering their stiffness properties to customize and reconfigure robot behavior without adding actuation complexity. It addresses three core gaps: the absence of systematic models for multi-material backbones with continuously varying stiffness centers and directional stiffness; the lack of mechanisms to selectively modulate directional/torsional stiffness in situ; and the limited actuation bandwidth and payload robustness of tendon-driven continuum robots (TDCRs). Early flexible manipulators typically used elastic backbones with constant stiffness properties. By modeling more general elastic structures with spatially varying stiffness, one can arrange stiffness properties to achieve a desired workspace using simple actuation inputs. We demonstrate this technique by shifting the workspace of a tendon-actuated flexible endoscope to improve performance in a simulated neurosurgical task based on a specific patient anatomy. Next, we explore how patterned phase-changing structures in silicone robots can selectively modulate stiffness. These thermally responsive structures can independently stiffen bending planes or torsion and change the robot’s reference shape (precurvature). Combining these spines with complex actuation, the same input can be mapped to different motion primitives, enabling workspace reconfiguration so that a small set of inputs can achieve a larger range of configurations. Finally, using a compact, low-cost, modular tendon actuation unit, we address some practical challenges in the actuation TDCRs, including the underdamped oscillations which limit their actuation bandwidth, and their robustness to unknown external loads. We demonstrate how antagonistic pretensioned tendons yield faster, more consistent responses and higher bandwidth without additional sensing or closed-loop control. Leveraging motor currents and displacements, we infer external loads and compensate their effect on shape without extra sensors. Together, these advancements provide a foundation for more capable and adaptable soft robotic systems.