Abstract
Robotic manipulation is a broad and complex field which has been studied for many decades, yet still with many open problems. Central to the field is the use of artificial tactile sensing to provide information from the contact interface such as contact force and location. Integrating high-resolution tactile sensors with robot manipulators is a particular challenge, with a desire to use the feedback afforded by the sensors to inform control policies for grasping and manipulation.This thesis documents the design and development of a biomimetic tactile sensor approximately the size of a human fingertip and its integration with an underactuated, anthropomorphic robotic hand. The resulting system is then applied to various grasping and manipulation tasks to investigate the interplay between the hand and sensors. This work includes numerous hardware advancements in tactile sensor design as well as in capturing and processing high-resolution tactile data efficiently. This thesis also presents novel methods for inferring contact force from tactile data via deep neural networks, as well as a study into how transfer learning methods may be used to improve accuracy.
This thesis aims to prove that the addition of high-resolution tactile sensing to robotic manipulators is crucial to the success of dexterous robotic manipulation and therefore vital in the advancement of introducing robots holistically into society. Ultimately, this work provides multiple contributions to the fields of tactile sensing and manipulation with underactuated, multi-fingered hands.
| Date of Award | 1 Oct 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Efi Psomopoulou (Supervisor) & Nathan F Lepora (Supervisor) |