Repository logo
 

Passivity-Based Hands Towards General-Purpose Robotic Manipulation


Type

Thesis

Change log

Authors

Abstract

Robotic manipulation in unstructured environments is a significant research challenge, where development contributes significantly to social robotics, industrial robotics and prosthetics. Despite a long history of research into artificial hands and significant development along the way, human-level manipulation remains unsolved. The current landscape of manipulation research reveals a dichotomy between the push for complex systems with diverse behaviours which can interact and adapt in novel environments, and the need for simplicity for practical realisation and applicability in the real world. The concept of embodied intelligence hypothesises that intelligent behaviours are largely due to physical interactions between a body and the environment. With the ability to embed intelligence into our manipulator design, complexity is more evenly distributed throughout the system, realising more capable systems with reduced control effort. This thesis explores the role of embodied intelligence within the hand, ways to exploit embodied intelligence and potentially ways to embed intelligence into our own designs. The concept of passivity-based design is presented. This seeks to extend embodied intelligence to passive hand behaviours, revealing design consideration and open challenges such as developing a benchmark to evaluate performance in general-purpose tasks. Furthermore, this thesis explores examples of embodied intelligence and how complex behaviours can emerge with passive design, such as with: an adaptive suction cup which utilises parallel chambers for multi-modal grasping; and a passive anthropomorphic hand which exhibits natural appearing behaviours and adaptive grasping through only wrist-driven interactions. The initial practical work investigates robotic manipulation in real-world applications with an emphasis on adaptability to task uncertainty and agility to changing task specifications. Applying the principles of passive design presented in this thesis, bespoke manipulation solutions are developed which address flexible assembly and bin-picking, both requiring robust and adaptive picking capabilities. The flexible assembly system is able to perform at a high level in a novel competition focused on agile manufacturing. The bin-picking system demonstrates wide coverage of objects seen in logistics and grocery warehouses. To investigate intelligence in more general-purpose systems, an anthropomorphic hand is developed. The design incorporates complex internal interactions with anatomical joint structure and tendon arrangement, allowing investigation into emergent passive behaviours. A framework for wrist-driven interactions is presented, which allows straightforward exploitation of passive behaviours such as grasping. Additionally, the hand is augmented with various features including joint stiffness control, proprioceptive sensors and tactile sensors, giving insights into iterative design as well as allowing further exploitation of self-structuring behaviours to predict future grasping outcomes from sequential interactions. The findings of this thesis demonstrate that intelligent passive behaviours have a significant role in the future of general-purpose robotic manipulation. Utilising the concept of passivity-based design, challenges such as the design of complex systems can potentially be addressed. Additionally, passivity-based design highlights the performance improvements gained with exploitation of behaviours in both existing and future hands through redundant wrist-driven interactions, environmental conditioning and passive information gain.

Description

Date

2022-09-19

Advisors

Iida, Fumiya

Keywords

Robotic design, Manipulation, Soft robotics, Robot interactions

Qualification

Doctor of Philosophy (PhD)

Awarding Institution

University of Cambridge
Sponsorship
EPSRC (2109088)
UKRI grant [RG99055] and Arm Ltd