Abstract

In this thesis, a new class of soft embedded sensors was conceptualized and three novel sensors were designed, fabricated, and tested for small force range soft robotic applications. The proposed soft sensors were consisted of a gelatin-graphite composite with piezoresistive characteristics. Principally,
the sensing elements of the proposed class of soft sensors were moldable into any shape and size; thus, were embeddable and scalable. The sensing elements were directly molded into soft
flexural structures so as to be embedded in the flexures. For each sensor, first a mechano-electrical phenomenological model for the exhibited piezoresistivity was proposed and validated experimentally. Afterwards, the sensors were subjected to a series of external forces to obtain calibration data. Given the complexity of the piezoresistivity and intrinsic large deformation of the soft bodies
and sensing element, learning-based calibration approach were investigated. To compensate for ratedependency and hysteresis effects on sensor readings in calibration, rate-dependent features were selected for learning-based calibrations. Consequently, the first sensor of this research, i.e., one degree-of-freedom (1-DoF) force sensor, exhibited a force range of 0.035-0.82 N force measurement range with a mean-absolute-error (MAE) of 3.7% and a resolution of 4% of full-range. The second sensor, i.e., 3-DoF had a measurement range of up to 0.3 N with an MAE of 0.005 N and
a resolution of 0.003 N. The third sensor, 6-DoF force-torque sensor, had a force range of up to 110 mN with an MAE of 7.4±6.5 mN and resolution of 1 mN and a torque range of 6.8 mNm
with an MAE of 0.24 mNm. Comparison with the state-of-the-art and functional requirements of intraluminal procedures showed that the the proposed sensors were fairly compatible with the requirement and showed improvement of the state of the art. The major contribution of this research was to propose a scalable sensing principle that could adapt its shape to the shape of the host body, e.g., flexural robots. Moreover, this research showed nonlinear learning-based calibration is a fitting solution to overcome limitations of the state-of-the-art in using soft elastomeric sensors.