UC Santa Barbara

When compared with traditional robotic manipulators that use discreterigid links and discrete joints, the sensing of soft robotic manipulators is inaccurate and their control is clumsy. This is because that the orientation of a traditional robotic arm can be known sufficiently by measuring relative rotation and translation of the links and joints subject to some initial condition. This could be done by implementing rotary encoders at every joint in a traditional rigid manipulator. Because of their continuum nature, a soft robotic arm’s orientation can only be known exactly if an infinite number were to be used. Therfore, the orientation of soft robotic manipulators can only be practically described when applying a set of simplifying assumptions such as piecewise-constant curvature. There are many ways to obtain information that can be used to describe the curvature of a soft robot; Motion capture is a prevailing technique in research but require that the robot is confined to spaces where motion capture can be used. Since this is not possible in many promising applications of soft robotics, including but not limited to minimally invasive surgery and disaster site navigation, many are looking to other ways to make robots capable of proprioception without using external sensors. One option is incorporating fiber optic sensing into soft robots. Fiber is flexible and a staggering amount of information about a system can be determined by reading changes in wavelength, phase shift, etc, which makes it an excellent solution for soft systems. The problem with incorporating fiber optics into soft robotics is that equipment and technologies such as optical spectrum analyzers, interrogators, fiber Bragg gratings are prohibitively expensive. Some work has been done showcasing that the optical intensity transmitted through a fiber can be correlated with the deformation that the fiber is experiencing (also known as macrobend loss or fiber optic intensity modulation), which can be read with much more accessible equipment. While promising, existing research only uses this technique to measure axial strain. We propose extending the method to determine arbitrary shapes of a continuum arm. We believe the power of data-driven methods will allow us to glean enough information from bending loss to achieve shape reconstruction currently only possible with high-cost optical spectrum analyzers. Specifically, this work aims to use a data-driven method to link the pose of a flexible arm with the intensity outputs of a number of fiber optic sensors attached to its body, using motion capture as ground truth. This work serves as the groundwork necessary for establishing a new low-cost optical sensing technique for soft robotics that could potentially allow important advancements in sensing and thus control.