Design, Modelling, and Fabrication of a Robotic Hand with Embodied Human-Like Compliance and Tactile Sensing

Abstract

This research aims to design and develop a new generation of robotic hands with human-like compliance and sensing. Our goal in the ongoing project is to close the gap between biomechanical studies and the design of robotic hands. Sensing and interpreting of contact forces, which are essential for precisely controlling interactions with the environment, are challenging for current robotic manipulation systems. Because it offers information that remote sensors like traditional cameras or lidars cannot, tactile sensing is suitable for manipulating robots. In unstructured situations, where the robot's internal image of the manipulated objects is unclear, touch is very important. To enable precise control of the perception of an object's physical attributes. This proposal will demonstrate the sensorization of 3D-printed soft-rigid robotic grippers. The proposed sensor types are the DIGIT sensor from Facebook and the GelSight sensor. By applying the sensing capability, the gripper can detect the size, roughness, contour, texture, pressing force, and other aspects of an object. Robot perception capabilities, especially those which utilize a tactile sensor to interpret the interaction between the controlled objects and the robotic manipulator, are essential for the appropriate automation of complicated human-like manipulation activities. In this work, different designs and prototypes of harvesting robots are presented. In addition, the presented harvesting grippers are used in farming with capabilities of plucking and delivery of pesticides by employing a sprayer and a liquid injector. Recognizing the roughness of the fruit and the fruit type is also present in this thesis work by utilizing Digit sensors which are vision-based sensors that capture the texture of the touched object. Interpreting the maps and images from the Digit sensor into a measurement of forces using the convolutional neural network CNN is also considered in the present work. SoRoSim, a MATLAB toolbox based on the geometrically exact variable strain (GVS) concept, is also used to analyze the fingers as well the trajectory of each finger statically and dynamically. Tendon wires are used to actuate and grip different objects in this work.

Date of Award	Aug 2023
Original language	American English
Supervisor	Irfan Hussain (Supervisor)