I am a PhD candidate and Graduate Research Assistant at
Missouri University of Science and
Technology, USA, under the
supervision of Dr. Sarangapani. My research explores the fascinating
intersection of reinforcement deep learning and safety control
mechanisms within nonlinear systems operating under multi-tasking
environments. The potential applications of my study stretch across
a broad spectrum including but not limited to robotics, mobile
robots, and both unmanned aerial and ground vehicles.
Research Interests:
Reinforcement Learning, Optimal-control, Deep Neural-Networks, Applications of Artificial Intelligence in Nonlinear-systems, Machine Learning Applications in Nonlinear-systems and Control, Robotics, Autonomous Vehicles, Human-Robot Interactions, Safety and Security for Nonlinear-systems, Microgrids, Smart Grids.
About me
I earned my M-Tech degree in Cyber-physical Systems from the
prestigious IIT Jodhpur, India, during which time I honed my skills
in problem formulation and experimentation using tools like
MATLAB, DSpace, and Opal RT. My thesis focused on the
development of controllers for power electronic converters and
microgrids, employing the Integral Sliding Mode Control approach
to mitigate ripples in power electronic converters. This work was
generously funded by DST.
During my time at IIT Jodhpur, India, I undertook several exciting
projects in the field of Cyber Physical Systems and Autonomous
Cars using image-based inputs (OpenCV, YOLO V3).
Motivated by my deep-rooted passion for research, I moved to
the Missouri University of Science and Technology, USA, to delve
into Deep Learning-based control for nonlinear systems in
multitasking environments. My work here largely revolves around
safety and control of robots and unmanned vehicles (ground and
aerial) with human interaction. I’ve used MATLAB for simulations
and provided mathematical proofs to support the proposed theory.
My research is currently funded by the Army Research Office,
Office of Naval Research, and the Intelligent Systems Centre.
My academic journey has been enriched by my contributions to
several international conferences and journals. My first-authored
works, which have received over 250 citations, cover a broad range
of research areas such as Cyber Physical Systems, Nonlinear
Control, Safety, Machine Learning, Deep Learning, Lifelong
Learning, and Robotics
Looking forward to further explorations and collaborations in this
exciting world of tech!
Institution: Missouri University of Science and Technology,
Rolla, USA
Tenure: Aug 2023- Present
I am deeply involved in a project that focus on the development of Human interaction with robots with safety in multi-environment scenarios using online deep continual learning.
Fig.1 - Formation control for leader and followers using deep continual learning optimal tracking based control in multi-environment scenario.
Institution:** Missouri University of Science and Technology,
Rolla, USA
Tenure:** Aug 2022 - July 2023
In this period, I contributed significantly to a project named
”Deep Continual Optimal Reinforcement Learning (DCORL) for
Tracking Control,” which aimed at nonlinear control systems. In this
project, we placed a significant emphasis on safety, which was
achieved through the implementation of time-varying constraints.
The system was designed to function seamlessly in multitasking
environments and found applicability in various domains like n-link
robot manipulators, mobile robots, unmanned aerial vehicles, and
unmanned surface vehicles.
Responsibilities and Achievements:**
Developed a nonlinear controller leveraging online Optimal
Deep learning with a continual learning feature to support operation
in environments that required the execution of multiple tasks
concurrently while minimizing cost and ensuring safety.
The project, supported by the US Naval Research Office and
the Army Research Office, culminated within the proposed timeline.
The outcomes were confirmed through rigorous testing and
simulation results, leading to several publications in reputable
journals and conferences.
Key aspects of this project encompassed the creation of a novel
adaptive tracking Deep Continual Optimal Reinforcement Learning
(DCORL) control scheme for nonlinear systems. This was achieved
by employing Singular Value Decomposition technique that
enhances the controller’s performance.
We incorporated continual learning techniques in an online
manner to devise a perpetually adaptable control scheme. This
continually learning controller can adjust to changes in system
dynamics, such as variations in mass or external disturbances,
ensuring optimal control even under changing conditions.
Our research evidenced that the proposed online DCORL-based
control scheme adeptly adapts to the dynamic shifts in the
environment, system, or path without significant loss of previously
learned information.
The stability of our developed control scheme was assured via
Lyapunov-based analysis, emphasizing the robustness and reliability
of our approach.
Fig.2 - BLF-preventing-violation-of-constraints
Fig.3 - Formation control Lifelong learning with barrier first TrajectoryFig.4 - Formation structure change in multi-environment scenarios.
Institution: Missouri University of Science and Technology,
Rolla, USA
Tenure: Aug 2021 - Aug 2022
During my tenure, I was deeply involved in a project that
focused on the development of an online Adaptive Lifelong Deep
Learning (LDL) for tracking control in nonlinear control systems.
This system was designed to function effectively in multitasking
environments and found applications in n-link robot manipulators,
mobile robots, and unmanned surface vehicles.
Responsibilities and Achievements:
Designed a nonlinear controller that leveraged Lifelong Continual
Deep learning to facilitate operation in multitasking environments.
This project, generously funded by the US Naval Research Office
and the Army Research Office, was successfully completed within
the proposed timeframe, and its results were demonstrated via tests
and simulations. These results further contributed to multiple
publications in renowned journals and conferences.
Key aspects of this project included the development of a new
adaptive tracking LCDL control scheme for nonlinear systems; we
successfully mitigated the vanishing gradient problem and
significantly enhanced the performance.
We also explored the use of continual learning techniques in an
online manner, thereby creating a continuously adaptable control
scheme capable of adjusting to changes in system dynamics, such as
shifts in mass, changes in the environment, or external disturbances.
Our findings demonstrated that the proposed online Continual
Learning-based control scheme was adept at adapting to dynamic
changes in the environment, path, or system without losing
significant previously learned information.
Crucially, the stability of our control scheme was guaranteed
through Lyapunov-based analysis, reinforcing the effectiveness and
dependability of our approach.
Fig.5 - Leader-follower-arc-shaped-trajectory
Fig.6 - Circular formation control.
Fig.7 - Tracking in multi environment with different path.
Institution: Missouri University of Science and Technology,
Rolla, USA
Tenure: Aug 2021 - Aug 2022
During my tenure, I was deeply involved in a project that
focused on the development of an Online safe adaptive Lifelong
Deep Learning (LDL) for tracking control in nonlinear control
systems. This system was designed to function effectively in
multitasking environments and found applications in n-link robot
manipulators, mobile robots, and unmanned surface vehicles. Safety
assurance was was accomplished through the implementation of
time-varying barriers. These dynamic barriers allowed us to securely
control system parameters and adapt to changing circumstances,
effectively minimizing potential hazards and enhancing the overall
reliability of the system.
Responsibilities and Achievements:
Designed a safe nonlinear controller that leveraged Lifelong Deep
learning to facilitate operation in multitasking environments.
This project, generously funded by the US Naval Research Office
and the Army Research Office, was successfully completed within
the proposed timeframe, and its results were demonstrated via tests
and simulations. These results further contributed to multiple
publications in renowned journals and conferences
Key aspects of this project included the development of a new
adaptive tracking LDL control scheme for nonlinear systems using the Singular
Value Decomposition (SVD) of the gradients. By
implementing the SVD approach on the Neural Network gradient,
we successfully mitigated the vanishing gradient problem and
significantly enhanced the controller’s performance.
We also explored the use of lifelong learning techniques in an
online manner, thereby creating a continuously adaptable control
scheme capable of adjusting to changes in system dynamics, such as
shifts in mass, changes in environment, or external disturbances.
Our findings demonstrated that the proposed online Lifelong
Learning-based control scheme was adept at adapting to dynamic
changes in the environment, path, or system without losing
significant previously learned information.
Crucially, the stability of our control scheme was guaranteed
through Lyapunov-based analysis, reinforcing the effectiveness and
dependability of our approach.
Fig.8 - Formation control and obstacle avoidance.
Fig.9 - UAV tracking in multienvironment and different path using Deep continual optimal tracking.
Fig.10 - Formation control using Continual deep optimal tracking.
Institution: : Indian Institute of Technology, Jodhpur
Tenure: July 2020 - July 2021
I played a significant role in a project, generously funded by
DST India, which was oriented towards the design of a nonlinear
integral sliding mode control for ripple mitigation in microgrids
Responsibilities:
Designed a nonlinear controller and provided mathematical proofs
for controller stability.
Designed necessary hardware for the project.
Ensured seamless software-hardware integration using DSPACE
and Opal RT.
Wrote various control algorithms using MATLAB and Keil
Embedded Development Tools for Arm, Cortex-M
Verified and validated the results
The project’s primary objective was to design a controller for
microgrids, aiming to reduce ripples, thereby enhancing performance
and improving robustness.
Institution: Indian Institute of Technology (IIT), Jodhpur
Tenure: Oct 2019 - July 2020
I played an instrumental role in a project aimed at Smart Field
Monitoring using Cyber-Physical Systems.
Responsibilities:
Developed an intelligent surveillance wireless sensor network
system utilizing ToxTrac and NS2. Ensured smooth software design
and implementation processes
Authored various algorithms using Network Simulator 2.
Conducted thorough validation and verification of the results
The project’s primary goal was to develop a cyber-physical
system-based approach for smart agriculture monitoring. This
innovative approach was designed to prevent extensive food loss and
environmental damage typically caused by rodents.
Institution: Indian Institute of Technology, Jodhpur.
Tenure: Oct 2019 - July 2020
I played a pivotal role in a project centered around the Design
and Implementation of a Real-Time Autonomous Car using Image
Processing and Internet of Things (IoT).
Responsibilities:
Developed an object detection algorithm leveraging both YOLO
(You Only Look Once) and Canny Edge Detection methodologies.
Assured seamless design and implementation of the associated
software. Conducted extensive verification and validation of the
outcomes. The project’s overarching aim was to create a fully
functioning autonomous car using image processing and IoT
technologies.
Image processing was utilized for real-time object and obstacle
detection, which is crucial for the car’s safe navigation. YOLO, a
real-time object detection system, was combined with Canny Edge
Detection, an algorithm used to detect a wide range of edges in
images, to create a comprehensive, effective detection system.
The IoT component of the project focused on allowing the
autonomous car to connect and exchange data with other devices
and systems over the internet. This data exchange capability is
crucial for updating the car’s operational parameters, reporting
status, and receiving instructions.
In this way, the autonomous car could react to its environment
effectively, navigating safely while interacting with a potentially
global network of information and control systems.
CONFERENCES ATTENDED
International Conference on Smart Electronics and Communication
(ICOSEC 2020), India
Third international conference on smart systems and inventive
technology (ICSSIT 2020), India.
6th IFAC Conference on Intelligent Control and Automation
Sciences ICONS 2022, Cluj-Napoca, Romania.
American Control Conference (ACC 2023), San Diego,
California, USA
7th IEEE Conference on Control Technology and Applications
(CCTA) 2023, Bridgetown, Barbados.
Strengths and Technical skills
MATLAB: Formulated and implemented the adaptive control
and estimation algorithms, published in American Control
Conferences and Journals.
Python: Devised learning based algorithms (Deep Learning and
Reinforcement Learning using Image data for autonomous vehicles)
published in conferences.
Robot Operating System (ROS): I have developed a working
knowledge of the Robot Operating System (ROS). This framework
allows me to efficiently create complex and robust robot behavior
across a wide variety of robotic platforms.
Network simulator: Implemented it on smart field monitoring
and in vehicular ADHOC networks, published in conferences.
dSPACE/HIL (Power electronics): Used to integrate Simscape
Electrical™ models in a dSPACE hardware-in-the-loop (HIL)
environment for testing microgrid converters
C++: Devised a working knowledge of C++, that helps in easier implementation of hardware integration with the software.
Personal Attributes:
Collaborative Mindset: My experience spans collaborating with
diverse project groups worldwide, leading to significant publications
and rich cross-cultural experiences. I value the diverse perspectives
brought by team members and find the collective intelligence
enhances the quality and innovation in our work.
Curiosity and Enthusiasm: I possess an unquenchable thirst for
knowledge, continually striving to learn new topics through diligent
self-study and dedicated efforts. I am driven by the power of
knowledge and the potential of applying it to solve real-world
challenges.
Self-driven and Tenacious: Originating from a modest
background in a small town, my journey so far has been fuelled by
unwavering determination and self-motivation. I have overcome
obstacles and embraced opportunities, culminating in my current
position in the field of research and technology. This journey has
instilled in me a resilience and a drive to strive for excellence,
regardless of where I am or where I started.
