Design of a 3D-printed soft robotic hand with distributed tactile sensing for multi-grasp object identification

Thesis

Tactile object identification is essential in environments where vision is occluded or when intrinsic object properties such as weight or stiffness need to be discriminated between. The robotic approach to this task has traditionally been to use rigid-bodied robots equipped with complex control schemes to explore different objects. However, whilst varying degrees of success have been demonstrated, these approaches are limited in their generalisability due to the complexity of the control schemes required to facilitate safe interactions with diverse objects. In this regard, Soft Robotics has garnered increased attention in the past decade due to the ability to exploit Morphological Computation through the agent's body to simplify the task by conforming naturally to the geometry of objects being explored. This exists as a paradigm shift in the design of robots since Soft Robotics seeks to take inspiration from biological solutions and embody adaptability in order to interact with the environment rather than relying on centralised computation.

In this thesis, we formulate, simplify, and solve an object identification task using Soft Robotic principles. We design an anthropomorphic hand that has human-like range of motion and compliance in the actuation and sensing. The range of motion is validated through the Feix GRASP taxonomy and the Kapandji Thumb Opposition test. The hand is monolithically fabricated using multi-material 3D printing to enable the exploitation of different material properties within the same body and limit variability between samples. The hand's compliance facilitates adaptable grasping of a wide range of objects and features integrated distributed tactile sensing. We emulate the human approach of integrating information from multiple contacts and grasps of objects to discriminate between them. Two bespoke neural networks are designed to extract patterns from both the tactile data and the relationships between grasps to facilitate high classification accuracy.

Authors: Shorthose, O

• DOI: 10.5287/ora-jn2nzoa8d
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: robotic hands; object identification; 3D printing; tactile sensing; soft robotics
• Subjects: Soft Robotics; Tactile sensors; Object Identification; Engineering