Type: dissertation
Abstract: Tactile perception is crucial for dexterous object manipulation in humans, and its adaptation in robotics is crucial for safe and effective interaction in unstructured environments.Despite significant progress, achieving sufficient sensitivity, robust and multimodal tactile sensing while ensuring integrability and scalability remains a challenge.This thesis explores the development of a biomimetic, tactile-sensitive artificial skin based on Fiber Bragg gratings (FBGs) to replicate human perceptual capabilities on robots.The research begins with fabrication of custom single-mode ZEONEX polymer optical fibers (POFs) with an in-house built POF drawing tower.The fabricated POFs offer superior flexibility, fracture toughness and strain sensitivity compared to silica fibers.Allowing seamless integration into soft, conformable skins on complex surfaces such as robotic fingers.First, an artificial skin with an array of four FBGs was developed for high-precision contact force measurement and localization on fingertips.The artificial skin withstood pressures up to 955 kPa without damage.Assistance by a random forest algorithm, the skin measured and located forces with mean absolute errors (MAEs) of 0.07 N and 0.24 mm, respectively, outperforming human tactile spatial acuity.Second, the capability to recognize object hardness was investigated.An artificial skin with four embedded FBGs captured unique strain signatures when pressed against objects of different hardness.An artificial neural network (ANN) processed these signatures, achieving a force estimation correlation of 0.96 and hardness classification accuracy of 92%, demonstrating excellent capability in simultaneous force estimation and hardness recognition capabilities.Next, an artificial skin with three-FBGs array and a biomimetic fingerprint-like pattern was developed for multidirectional force and vibration sensing.Normal force of up to 10 N and shear forces of up to 4 N were tested on the skin.A self-developed multi-input-multi-output convolutional neural network (MIMO-CNN) estimated applied forces with correlation of 0.96 and distinguished between 20 fabric samples with 94% accuracy via slip induced vibrations.Finally, a skin with thermal-based object recognition capability was developed.It utilized cobalt-doped photothermal fiber for heating and FBGs for temperature sensing.By monitoring hear transfer upon contact, an ANN classified metal, glass and wood with 100% accuracy within 5 seconds providing a complementary recognition modality to physical sensing. IIThis thesis establishes a comprehensive framework for polymer fiber Bragg grating (POFBG)based artificial skin, demonstrating exceptional performance in force localization, hardness discrimination, texture recognition and thermal material identification.This work also validates the combination of POFs and machine learning as a viable and scalable platform for advanced robotic tactile perception.
Author: Leong, Chern Yang
Language: English
Topic: Advanced Fiber Optic Sensors
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7217682437