Embedded Files
Abstract
Aquatic mammals, such as pinnipeds, utilize their whiskers to detect and discriminate objects and analyze water movements, inspiring the development of robotic whiskers for sensing contacts, surfaces, and water flows. We present the design and application of underwater whisker sensors based on Fiber Bragg Grating (FBG) technology. These passive whiskers are mounted along the robot’s exterior to sense its surroundings through light, non-intrusive contacts. For contact tracking, we employ a sim-to-real learning framework, which involves extensive data collection in simulation followed by a sim-to-real calibration process to transfer the model trained in simulation to the real world. Experiments with whiskers immersed in water indicate that our approach can track contact points with an accuracy of <2 mm, without requiring precise robot proprioception. We demonstrate that the approach also generalizes to unseen objects.
A) Robot arm instrumented with a whisker sensor array in a crowded fish tank. B) A close view of the whisker array in water. C) Predicted contact positions when the whisker array sweeps over 3 different objects.
Underwater robots can benefit from whiskers in cluttered scenarios. Robots such as OceanOne sometimes operate in murky water and have even become stuck on underwater obstacles. If cameras are not able to see where the contact occurs, freeing the robot may require external assistance. Recently, commercial counterparts to OceanOne have been developed (e.g. Honda underwater ROV) and these too may benefit from contact or proximity sensing for safe operation.
Collecting large-scale data data on contact positions from the real world is a challenging and laborious task. We therefore created a digital twin of our whisker sensor in the MuJoCo simulator, which approximates the real whisker dynamics.
The real-world data includes FBG wavelength shifts and camera images of the deflected whiskers, including contact locations---but does not provide whisker base moments directly. Accordingly, as shown in above figure, we train a transformer decoder using pair data of contact positions and base moments. In this section, we first introduce the data collection process and then describe the simulation model, which includes some adjustments to simplify computation while also accurately capturing the mapping from contact locations to whisker base moments. Finally, we present a sim-to-real calibration system designed to convert bending strains and associated wavelengh shifts to orthogonal bending moments at the base of the needle, for reconciliation with the results of simulation.
Page updated
Report abuse