Continuum Deformation Based Autonomy - Approaches to Multi-Agent Systems

Abstract

Unmanned Aerial Vehicles (UAVs) have gained widespread applications due to their versatility, agility, vertical takeoff and landing capabilities, and their relatively simple mechanical design. UAVs demonstrate an ability to execute tasks such as payload delivery and surveillance with a speed unattainable by a single UAV, particularly when operating as teams of UAVs. However, such team tasks increase the system's complexity and present significant challenges in ensuring safe team coordination and control. Reliability is another critical aspect of multi-agent system (MAS) coordination as it guarantees the MASs accurate and consistent performance. If one agent in the MAS fails or behaves unexpectedly, it can negatively impact the performance and effectiveness of the entire MAS. Therefore, it is important to design and implement algorithms for multi-agent systems (MASs) with a high level of reliability to ensure that they can operate safely and move smoothly in the presence of agent failures or lack of communication with other MASs moving in a shared motion space. Motivated to address the critical issue of safe team coordination, this dissertation advances previous research by presenting an innovative approach centered on the theory of ``Continuum Deformation", a continuum-mechanics based theory for the evolution of a Multi-Agent System (MAS). Within this theoretical framework, a MAS has the potential for expansion and compression while maintaining its formation. This approach enhances the fault tolerance, reconfiguration capabilities, and overall efficiency of the system. Importantly, it provides effective solutions to the pressing issues of safe team coordination, inter-agent collision avoidance, and maintaining the desired formation. This dissertation aims to further current research in this domain through presenting novel framework that hold promising societal benefits and future research directions. This thesis offers five specific contributions to enable the deployment of teams of UAVs in various missions in safe manner. First, we present a novel approach for analyzing and visualizing deformations in 2-D and 3-D motion space using a team of UAVs, while guaranteeing inter-UAV collision avoidance. We propose to treat each UAV as particles of a deformable body and apply the principles of continuum mechanics. We introduce a new role for a team of UAVs in the education sector as “teachers”, which provides an excellent opportunity to practice theoretical concepts of mechanics in an engaging manner. This work supports the use of robotics in delivering lectures in the education sector, generating increased interest and exposing students to multiple domains simultaneously. Second, we explore the challenge of safe and optimal deformation of a large-scale multi-agent system (MAS) while tracking a desired trajectory. We propose an innovative approach using a hierarchical leader-follower configuration similar to a Fully Connected Neural Network (FCNN) architecture. This framework allows us to model the coordination problem as a quadratic programming problem, where imposing bounds on the decision variables ensures inter-agent collision avoidance. We demonstrate the success of the resulting optimization problem through a numerical simulation on a large-scale quadcopter team following a helix trajectory. Third, building upon the previous two works, we introduce an additional decision variable for safe coordination of a large-scale multi-agent system (MAS) with ``local deformation" capabilities. The strategy builds on the previous hierarchical leader-follower framework and frames multi-agent coordination as a Deep Neural Network (DNN) optimization problem. The deep neural network plans the desired deformation based on the agents desired positions. This approach adjusts the weights of the neural network, where the weights are constrained by lower and upper bounds, to prevent inter-agent collisions. The method is validated through a simulation of a team of quadcopters following a desired elliptical trajectory. Fourth, we develop a physics-inspired algorithm to establish safe and resilient operation of a Multi-Unmanned Aerial Systems~(UAS) (MUS) in the presence of disturbances and unforeseen UAS failure. We propose the algorithm based on two operating modes: Homogeneous Deformation Mode (HDM), and Failure Resilient Mode (FRM). The transitions between HDM and FRM are formally specified using Cooperative Localization (CL) to quantify UAS tracking error, and detect anomalous conditions due to UAS failures. This capability will support resilient operation of MUS moving collectively in a shared environment. The main goal of this work is to develop a physics-inspired algorithm ensuring safety of large-scale UAS coordination in the presence of unexpected actuation failure/s. Fifth, we emphasize the importance of reliability in multi-agent system coordination to ensure consistent performance and prevent negative impacts from agent failures. We introduce a unique navigation model, applicable in ideal fluid-flow situations, which categorizes agents into cooperative (non-singular) and non-cooperative (singular) agents. This model enables cooperative agents to slide along streamlines, safely encompassing non-cooperative agents within a shared motion space. The proposed model's effectiveness has been validated through a series of flight experiments using teams of crazyflie quadcopters. The use of multi-agent systems (MASs) for various applications and well-defined framework offered in this dissertation serve as a foundation for future research in the domain of Multi-Agent Systems (MAS). This dissertation presents problem definitions, solution techniques and experimental results based on continuum deformation for five unique tasks. Drawing from the essential conclusions and knowledge acquired through results, signifies a leap forward in the field of multi-agent systems (MASs) with promising societal benefits. Furthermore, exploring new applications and domains for the continuum deformation based autonomy framework will advance the state of the art in multi-agent systems (MASs) and provide valuable insights for addressing complex problems.