I would like to thank the leaders of the Department of Computer Science and Technology during my Ph.D. To my knowledge, no part of the work reported in this thesis has been presented for the award of any degree at any other institution.
Cyber-Physical Systems
Internet of Things
An embedded device equipped with sensors, i.e. thing, must first be identified using addressing techniques such as IPv6. The things at the blade end, sample the data and upload it to the cloud platform.
Networked Robotics
Although this provides a simple cross-communication architecture, client-server-based communication is challenging to scale [100].
Decentralized CPS
However, decentralized systems only need to connect a node to any of the nodes in the network, making them scalable without affecting performance. With the increase in the number of devices and robots, research into decentralized CPS offers promising opportunities for new applications.
Research Challenges in Decentralized CPSs
However, some key questions that need to be answered in such scenarios are what and how to share. Choosing the best solutions (algorithms, heuristics, mechanism, etc.): Finding the best solution for a given problem is the essence of any intelligent CPS.
Mobile Agents
The mobile agent then resides in the new node until a connection to the next node is available. A mobile agent can add new behaviors in the form of payloads and can also adapt to different situations.
Contributions of the Thesis
Choose the best solution in a dCPS: The previous contribution assumed that the solutions available within the network are distinct and the best. This contribution proposes an immunology-inspired mechanism to find the best solutions for problems distributed over the network system.
Outline of the Thesis
A Multi-Agent System (MAS) is a paradigm that can meet the requirements of managing different processes within a CPS. In this chapter, Tartarus is transformed into a software tool that provides distinctive features that allow cyber agents to easily access and control the real physical devices.
Bridging Frameworks
Tartarus: A Multi-Agent Platform
Architecture
TartarusThread Handler (TTH): TheTTH acts as a controller for managing multiple threads of execution within the Tartarus platform. TartarusPlugin Channel (TPC): TheTPC forms the crucial component that extends the CPS-based functionalities of the Tartarus platform.
Features
It facilitates interfacing of the platform with entities or hardware controllers such as Pi and NXT robots. By modifying the payloads, one can try to change the behavior of the agent at runtime.
Tartarus for CPS
Tartarus: Real-World Application
Discussions
So Claus (or the human administrator) can request his nearest TAH to create mobile agents to guide Rudolph to the site of action. Furthermore, if the tasks involved are to be performed in a cooperative and synchronized manner, the same can be achieved with the help of a set of mobile agents [87].
Chapter Summary
The cloud extends to the user side and constitutes a group of distributed and decentralized computing nodes that form the edge of the network. Privacy: Most cloud servers are owned by multinational corporations such as Amazon, Google, Microsoft, Cisco, etc.
Background
A point location model does not provide facilities for interpolating or extrapolating moving object location data and is not accurate. In a trajectory location model, an estimate of the source and destination of a moving object is determined.
Decentralization using Mobile Agents
- Detection Mechanism
- Wearable and Acquisition Unit (WAU)
- Cyber Computing Unit (CCU)
- Motion Vector
- User Interaction Unit (UIU)
When a person enters a zone within the corridor, the BLE receiver of the Pinode within that zone detects his/her presence in that zone. Once the mobile agent enters this Pi-node, it executes its code and eventually ends up in the Pin-node of the zone the person is currently in.
Experiments and Results
- Data Acquisition
- Query Processing
- Scenario 1: Conventional Cloud approach
- Scenario 2: Proposed dCPS approach
- Comparison of Scenario 1 with Scenario 2
As the BLE tag carriers (mobile agents) move through the building (network), all relevant data within the Pi node (such as timestamp, UUID, motion vectors, etc.) is sent directly to the cloud. In both cases, data transfer costs were recorded in terms of the number of times the Pi nodes connected to the cloud.
Chapter Summary
In the domain of distributed computing, mutual exclusion of shared resources (MER) is called the classical comparison problem for resolving resource contention [151]. In a token-based approach, a node with a unique token can access shared resources while others have to wait for the token to arrive.
Work in Brief
In the field of robotics, the problem of synchronization is exacerbated, as the time required for a robot to perform a given task can vary due to several environmental factors. The Task Execution Order Problem (TEOP) among multiple robots and the inherent goals for realizing CPS are discussed in the following sections.
Preliminaries
Constituents of the proposed CPS
Agents: A set of mobile agentsµ= {µ1, µ2,µ2,…,µm|m≥1}, such that each mobile agentµi∈µ, carries as its payload the programs of its associated tasks. Job: A Job Ji is a collection of tasks in T together with the associated set of resources inΨ, to be executed by the robots inRand and constitutes the basic input to the system.
System Specifications
The Task Execution Ordering Problem
Inherent Objectives
Objective 1 basically makes robots in R coordinate their implementations in a pipelined fashion. The second objective concerns the preservation of the time period of each stage in an asynchronous robotic pipeline.
The Proposed Mechanism
- Job Distributor
- Mobile Agent based Mechanism
- Asynchronous Execution Times
- On-the-fly Addition/Deletion of task(s)
- Mutual Exclusion for Concurrent Tasks
- Avoiding Deadlocks
It also includes the state information (SI) of the robots that can serve along with the other state the robots need to transit to. In the current decentralized and distributed CPS, a mobile agent is the only entity that has the code to execute a specific task.
Implementation
Thus, to show that the proposed system is deadlock-free, it is sufficient to prove that the graph is acyclic. Tartarus [163], a mobile agent platform was used for emulating the proposed mechanism for sequential and interdependent task execution.
Experiments and Results
- Emulation
- Comparison with a Centralized Controlled System
- Task Addition and Deletion
- Experiments on a Robotic Warehouse
- A mix of Sequential and Independent tasks within a job
- Modification of the Sequence of Tasks
- Removal of robots
However, as the number of tasks grows (more than 30), the throughput of the centralized system decreases rapidly because the average time to complete a task increases. This caused the length of the sequence in a job to vary at runtime.
Chapter Summary
The Adaptive Immune System
The paratope, which forms the antigen binding site of the antibody and the corresponding complementary shaped epitope on an antigen, facilitates antigen recognition. The following sections describe the significant contributions and provide a brief chapter-wise summary of the thesis.
Theories of Biological Immune System
Clonal Selection
Antibodies that can recognize the antigens bind and become activated resulting in a clone. Cloning and hypermutation together add diversity to the antibody repertoire of the adaptive immune system which in turn helps improve antigen recognition.
Idiotypic Network
The antibodies that are more effective in this recognition process eventually become memory cells that remain for a longer time in the biological being. These stimulations and suppressions finally result in proliferation or decay of the antibody molecules.
Danger Theory
Features
Adaptability: During the lifetime of the host body, the BIS generates responses to the attacks it has encountered in the past. Without the help of any external entity, the various components of the BIS organize themselves to the incoming foreign invasion.
Artificial Immune System
Shape-Spaces and Cross-Reactivity Threshold
An Ab is selected for anAg only if the complementary shape of the Epof thatAg lies within a small region of the shape space surrounding that of the Pt of Ab. As shown in Figure 5.5(b), the Ab can control two of the Ags that lie within its AR.
Affinity Measures
This small region, called the active region (AR), is characterized by a cross-reactivity threshold (ϵ) [43]. Briefly, the role of BIS is to generate, select and evolve the range of Abs best suited to contain antigenic attacks.
Applications
To maintain good solutions, their concentration or population is increased as in BIS. The work presented in the next chapter also uses the same principle to manage and control heterogeneous antibody populations.
Chapter Summary
166] have described some benefits of peer-to-peer (P2P) local knowledge sharing in such decentralized cyber-physical systems (dCPS). Sharing has been proven to be useful in improving the performance of the decentralized and distributed systems, as in [90].
Motivation
Deciding the metadynamics of the system so that good solutions are rewarded while bad ones are suppressed. In the next section, some basic mechanisms found in BIS are briefly introduced, followed by the proposed approach and the associated metadynamics.
Brief Survey
Although the presented problem of choosing the best solution to a problem in a decentralized and distributed way may seem naive, it poses a number of interesting challenges that are crucial in the realization of information exchange and mapping. In the immune system-inspired work discussed above, problems are represented as antigens, while solutions are formed by antibodies.
The Proposed Approach
Intelligent Packets Migration
In this scheme, each mobile agent, called Intelligent Packet (iPkt), determines its migration strategy based on the current state of the environment. Thus, based on the status of the nodes, the iPkts migrate to them, which is unlike broadcast-based methods.
The Proposed Mechanism
The signals also carry the antigenic problem instance FV with them and penetrate the adjacent nodes in the form of a diffuse signal gradient. In BIS, affinity is based on the degree of complementarity between the shapes of the paratopes and the epitopes.
Time Complexity Analysis
Experiments and Results
- Emulation
- Experiment on a 10-node physical network
- On-the-fly Addition of Solutions
- Experiments on a 50-node physical network
- Performance in a Dynamic Network
- Effect of Influx Rate
- Effect of Barrage Size
- Effect of Danger Signals
- Experiment on an IoT scenario
- Comparison with the non-sharing approach
- Experiments using Real-Robots
It can be noted that the population represents the trend in learning of the system that implements the proposed mechanism. As can be seen from the table, the results obtained validate the correctness of the proposed mechanism.
Chapter Summary
Behavioral decomposition
Lee [116] describes each of the low-level subtask-specific controllers as behavior primitives while the high-level controller that has learned to map the inputs to these subcontrollers was called the behavior arbiter. This repertoire is then used to evolve an ANN which maps the robot's sensory inputs to an appropriate primitive.
Embodied Evolution
Once the appropriate sub-controllers were found, a behavior arbiter was developed to delegate the sub-task to the best sub-controller based on the sensory inputs. Although the complex task was solved using their hierarchical controller, the behavioral primitives and the behavioral arbiter seem not to have been developed in an online and onboard manner.
Methodology
From Immunology to the Real World
The fitness function is task dependent and provided by the experimenter on an application basis. Like BIS, this algorithm is decentralized and distributed in nature and evolves sub-controllers on the fly.
The Proposed Algorithm
The Ctr of the best Ab is developed to produce adequate behavior within the associated AR. In this strategy, the controller weights are mutated using a Gaussian. a) The robot (b) The experimental arena.
Experiments
Scenarios
The sub-goals here include moving the robot in the direction of the light source and avoiding obstacles. However, when the light source was OFF, the robot had to learn to deflect these objects when it encountered them.
Results
Evolving a Single Robot Controller
Experiments using Real-Robots
This indicates that the robot has detected the next Ag within the AR of the same antibody. In this trial, the robot was randomly placed in a position facing the wall of the arena.
Architecture of Tartarus
A depiction of the Tartarus platforms configured as a CPS
The top-view of the implementation setup of the Cyber-Physical Sys-
The Pi based Tartarus host interfaced to the Webcam
The snapshot captured by the Webcam mounted on the Pi based
An Agent based Cyber-Physical System
Decentralized and Distributed LATS
A sample snapshot of the part of the database maintained at a Pi-node 44
BLE raw and filtered data
Graph showing Inter-Zonal movement for a single BLE tag bearer . 51
Traffic flow for centralized cloud based and mobile agent based systems 54
Pipelined execution of a set of sequential and interdependent tasks