Shape Reconstruction Using Soft, Stretchable Sensors With Embedded Strain Sensors and Inertial Measurement Units

In this article, we present a custom soft sensor for performing shape reconstruction of continuum structures. One major drawback of camera-based motion capture systems is the need for a structured, instrumented environment. Outside of these curated spaces, there are many applications that would benefit from having a general-purpose, camera-less sensing solution for proprioception. Our approach is a sensor that performs self-shape reconstruction with no need for external data collection, structured environments, or model training. The sensor body is made of soft silicone and embeds a series of custom printed circuit boards (PCBs) with nine-axis IMUs and liquid metal strain sensors spanning between them, and supports up to 100 Hz sample rate for all sensors. We use spherical interpolation with the strain and IMU orientation measurements to sequentially reconstruct the sensor geometry along the chain of segments. We validated the sensor design and accompanying reconstruction technique with dynamic bending experiments on a custom, motorized testbed. Using optical motion tracking to ground truth the reconstruction estimates, the average Euclidean distance errors over the body of the 325 mm long, five-node sensor ranged from 5.07 to 9.38 mm. Additional cycle testing demonstrated that the sensor behaved consistently over longer durations, with low deviations in estimated positions when moved through the same positions.