Bio-inspired sensing and control strategies for aerial manipulation
File(s)
Author(s)
Xiao, Feng
Type
Thesis
Abstract
Aerial manipulation enables aerial robots to physically interact with environments. Enhancing aerial robots with manipulators allows them to perform tasks such as contact-based inspection, object pick-up, repairing, and building. However, small robots under 250g do not have the luxury of carrying additional manipulators due to their extremely limited payload. By reviewing the current research field and implementing two aerial manipulation systems, I found that most aerial robots use sensing and control systems based on a sensing-modelling-planning-action (SMPA) strategy, which required sophisticated sensors and powerful computers. These devices already took up most of the available payload, leaving insufficient payload for additional manipulators on small robots. Flying insects, despite their simple nervous system and tiny body, have superb navigation skills, and seem effortless when performing aerial manipulation tasks such as perching, building nest, and collecting food. By observing bees' behaviours, I found that complex tasks are achieved by combining different direct sensing-action (SA) pairs. This thesis took this new approach that uses a bio-inspired SA strategy to make the sensing and control system of small aerial robots more compact, consume less payload, and still provide enough autonomy. I designed a behaviour-based multi-sensory framework based on SA strategy and applied bees' sensing and control principles on robots. This framework uses multi-modal sensing and direct sensor-action links to form computationally simple behaviours, activates different behaviours based on different conditions, and prioritises the activated behaviours to achieve a complex task. Using this framework, the designed robots can use low-mass sensors, and the required computations can be done on small microcontrollers. Two robots powered by this system were validated in vertical surface interaction scenarios. With total take-off masses of merely 158g and 100g respectively, the robots can detect the unknown vertical surfaces, avoid collision, and contact surfaces with near-zero velocities. With a bio-inspired system that is compact and robust, this work establishes the fundamental sensing and control framework for lightweight aerial manipulations.
Version
Open Access
Date Issued
2022-02
Date Awarded
2022-10
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Kovac, Mirko
Sponsor
Imperial College London
Grant Number
EP/R009953/1
EP/N018494/1
EP/R026173/1
Publisher Department
Department of Aeronautics
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)