div.absolute { position: relative; top: 100px; right: 0;} Akhilesh Raj
This is me - Control Engineer

AKHILESH RAJ

Ph.D scholar with the Engineering Graduate Fellowship at Vanderbilt University.

A motivated researcher in the field of control of cyber physical systems and machine learning with hands on experience in problem solving - theory and application, using math and math-tools. The knowledge seeker and the team player in me helped gain experience in electronic assembly and Python coding.

Areas of Interest: Reinfrocement Learning, Control Systems, Distributed Estimation, Surrogate Modeling, Machine Learning.

Project: Composition of surrogate models.

Positons holding/ held: Advisory Board Member, Bayesian ways.LLP Lead Project Engineer, DropVault Tech. Pvt. Ltd.

About me

Having pursued my research career as an M‑Tech graduate from IIT Kanpur, I gained experience and expertise in problem formulation, solving and experimenting using various software and hardware tools like LABview and ROS. My M.Tech thesis work was the development of an Automated Guided Vehicle (AGV), which I did at IDEA Lab (IIT‑Kanpur). The Boeing Company funded project demanded automation of their warehouse vehicles for material handling using a minimal number of sensors. Therefore the hardware was designed and developed at a minimal cost and the sensors for object detection, obstacle avoidance, line following, tracking,recognition etc. were replaced with a camera. The simulations were performed using MATLAB and the backend support was provided using python with the added library support from OpenCV and Arduino Uno.

Following my Master’s degree, my urge towards research studies pulled me to the Missouri University of Science and Technology, Missouri, USA, to work in the field of control theory. My research was in distributed state estimation with the objective of estimating the position and velocity of an unknown moving target in real time. The algorithm was developed using adaptive observers and neural networks, and mathematical proofs were provided to support the proposed theory. Similarly, the simulations were performed using MATLAB, and were shown supporting the proposed theory. The project was funded by the Intelligent System Center and Dynamic Data Driven Application of the Air Force office for Scientific Research, USA.

I have 9 publications in various international conferences and journals, which has over 35 citations. These first authored works span a wide range of research areas which include distributed estimation, control, machine learning, computer vision etc. I also have abook Chapter related to learning and estimation published from IIT Kanpur in Computational Intelligence: Theories, Applicationsand Future Directions ‑ Volume II.

Currently I am an Engineering Graduate Fellow at the Vanderbilt University, Nashville, Tennessee, USA in the department of Electrical and Computer Engineering, working towards my PhD. in the area on control of cyber‑physical systems

Python
MATLAB
WebGME
LABVIEW
bash
HTML coding
C++

Work Experience

Argonne National Lab

Role: W.J Cody Associate

Period: May 2022 - August 2022

ARGO node resource management: At Argonne National Lab I worked as a W.J Cody associate as a part of UChicago Argonne summer internship program. I was employed in the Argo Project under the department of Mathematics and Computer Science. During the internship I developed a novel power optimization algorithm for high performance computers (HPCs) using a reinforcement learning approach leveraging the capabilities of Intel Running Average Power Limit (RAPL) technology. Power consumption of HPCs have been a topic of research since its development. The main bottleneck was to determine the relation between the performance and power which was pivotal for the control of power provided to a compute-node or a processor, without effecting its net performance. The execution time required for a standard STREAM application was used to evaluate the performance and the workload or the application heartbeats provided an instantaneous measure of the performance of a node. During my period of employment there at Argonne National Lab, I was able to come up with a reinforcement learning based algorithm that used a Proximal Policy Optimization (PPO) agent to learn a policy which optimized the power cap that was provided to the RAPL actuators. By performing the evaluation of the algorithm against the minimum and maximum performance configurations, it was shown that our results fell in the region of optimal performance.

Indian Institute of Science, Bangalore, India

Role: Project Associate

Period: March 2021 - July 2021

Intersection Management: This project in the Networked Control Systems Lab aimed at developing Intersection Management algorithms for connected and automated vehicles (CAV). The requirement of the project was:

  • Design decentralized algorithms for a CAV to determine its velocity such the passenger safety, collision avoidance and intersection safety will not be compromised. The non-linear optimization problem was formulated and simulated using Casadi.
  • Due to the increasing time complexity and power requirement in the non linear optimization method a Reinforcement learning based approach was on the development.
  • Parallelising the simulation program in python on multicore processors was one of my important accomplishments.
  • The hardware implementation of the theory was initiated on the m3pi pololu robots using raspberry pi and the hardware implementations were carried forward

Dropvault Tech. Pvt. Ltd.

Role: Lead Project Engineer

Period: October 2020 - February 2021

Secure package delivery system: The Bangalore based start up recruited for making a fully functioning prototype of their dream product.

  • Designed and developed a functional prototype of a secure package collection system, working from the concepts towards reality.
  • A Raspberry Pi based prototype model was built by integrating the chosen components. \item The back-end software for the functioning of the prototype was built using Python.
  • The main features include:
    • All weather hardware.
    • ID based package delivering and monitoring system.
    • App support.
  • The following python packages were used in the prototype
    • socket (conventional TCPIP)
    • socketio (event based TCPIP)
    • asyncio (asynchronous concurrent execution)
    • GPIO (Rpi Pin Control)
    • cv2 (Open Computer Vision)
    • Aiohttp (WebRTC)

Fig.1 - Prototype video from pitch deck (credits: source video)

Bayesian Ways LLP.

Role: Research Consultant

Period: August 2020 - Oct 2020

A short term challenge hosted by one of the upcoming startups in India. The challenge was to solve an optimization problem associated with optimal Operation Room scheduling in the hospitals. Accomplishments include:

  • Developed MATLAB and Python based programs for an optimal event scheduler.
  • The initial code development was done in MATLAB using GUROBI and MOSEK (licensed cvx solvers).
  • Parallelising the simulation program in python on multicore processors was one of my important accomplishments.
  • The solver for the event scheduler(a mixed integer problem), was then developed in Python using CVXPY and OR-TOOLS.

Missouri University of Science and Technology

Role: Research Associate

Period: January 2017 - December 2019

  • Worked on developing a distributed state estimation architecture for multi-agent systems with applications to target tracking applications.
  • The project, funded by Dynamic data driven applications of air force office of scientific research, demanded the detection and tracking of an enemy air-craft in practical scenarios.
  • Delivered the project on time with test and simulation results. The results were accounted as publications in well known journals and conferences. Briefing:
    • Estimation and control architecture to accurately determine the position as well as velocity.
    • The algorithm works even in the environment where target dynamics as well as the inputs are unknown.
    • Tested for a linear, non linear and event-triggered cases using MATLAB.
  • Responsibilities:
    • Propose new control theories/ techniques.
    • Prove the proposed theorem using mathematical tools like Lyapunov analysis.
    • Validate the results through simulations using MATLAB.
    • Perform control system experiments with Quanser interfaces using MATLAB/ LabVIEW and relate the inferences with the control theory.
    • Trained and mentored the disciplines of control system.

Fig.2 - Flocking using distributed sensing.

Indian Institute of Technology, Kanpur

Role: Project Engineer

Period: July 2016 - December 2016

  • Worked as a Project Engineer in a Boeing funded experiment in developing automated guided vehicles (AGV) capable of material handling.
  • Responsibilities.
    • Design the hardware.
    • Ensure smooth software hardware integration.
    • Coding various control algorithms using C++.
    • Validate the results.
  • The motive of the project was to replace majority of costly sensors that were being used in the industry, by using computer vision algorithms or cheap sensors.
  • Web cams were used to implement and test following algorithms:
    • Color line following.
    • Obstacle avoidance.
    • Object following.
    • Pattern Recognition.

Fig.4 - Demonstration of AGV.
Fig.3 - Demonstration of vision based object tracking in a random video.
Fig.5 - Demonstration of vision based object tracking in a random video.

Publications

[1] A. Raj, S. Perarnau, and A. Gokhale, “Performance-Aware Power Reduction in Exascale Computing: Leveraging Reinforcement Learning for Unified Control of Diverse Applications,” in 2023 Awaiting submitted to "38th IEEE International Parallel & Distributed Processing Symposium" [ bib ]
[2] A. Raj, S. Perarnau, and A. Gokhale, “Reinforcement-Learning Based Performance-Aware Power Reduction in Exascale Computing: Strategies for Application-Agnostic Agents,” in 2023 Submitted to "The International Conference for High Performance Computing, Networking, Storage, and Analysis" [under review] [ bib ]
[3] A. Raj, S. Perarnau, and A. Gokhale, “A Reinforcement Learning Approach for Performance-aware Reduction in Power Consumption of Data Center Compute Nodes,” in 2023 11th International Conference on Cloud Engineering [accepted for publication] [ bib ]
[4] A. Raj, S. Jagannathan, and T. Yucelen, “Event-triggered adaptive distributed state estimation by using active-passive sensor networks,” in 2019 American Control Conference (ACC), pp. 4695--4700, IEEE, 2019. [ bib ]
[5] N. K. Verma, P. Nama, G. Kumar, A. Siddhant, A. Raj, N. K. Dhar, A. Salour, et al., “Vision based object follower automated guided vehicle using compressive tracking and stereo-vision,” in 2015 IEEE Bombay Section Symposium (IBSS), pp. 1--6, IEEE, 2015. [ bib ]
[6] N. K. Verma, G. Kumar, A. Siddhant, P. Nama, A. Raj, A. Mustafa, N. K. Dhar, and A. Salour, “Vision based obstacle avoidance and recognition system,” in 2015 IEEE Workshop on Computational Intelligence: Theories, Applications and Future Directions (WCI), pp. 1--7, IEEE, 2015. [ bib ]
[7] A. Raj, A. Sivaraman, C. Bhowmick, and N. K. Verma, “Object tracking with movement prediction algorithms,” in 2016 11th International Conference on Industrial and Information Systems (ICIIS), pp. 285--290, IEEE, 2016. [ bib ]
[8] A. Raj, S. Jagannathan, and T. Yucelen, “Distributed adaptive state estimation and tracking by using active-passive sensor networks,” International Journal of Adaptive Control and Signal Processing, vol. 34, no. 3, pp. 330--353, 2020. [ bib ]
[9] A. Raj, S. Gupta, and N. K. Verma, “Face detection and recognition based on skin segmentation and cnn,” in 2016 11th International Conference on Industrial and Information Systems (ICIIS), pp. 54--59, IEEE, 2016. [ bib ]
[10] A. Raj, K. Gandhi, B. T. Nalla, and N. K. Verma, “Object detection and recognition using small labeled datasets,” in Computational Intelligence: Theories, Applications and Future Directions-Volume II, pp. 407--419, Springer, 2019. [ bib ]
[11] A. Raj, S. Jagannathan, and T. Yucelen, “Distributed state estimation by using active-passive sensor networks,” in 2019 American Control Conference (ACC), pp. 4689--4694, IEEE, 2019. [ bib ]
[12] A. Raj, S. Jagannathan, and T. Yucelen, “Distributed adaptive state estimation and tracking scheme for nonlinear systems using active passive sensor networks,” in 2020 American Control Conference (ACC), pp. 2587--2592, IEEE, 2020. [ bib ]

Contact me

Phone: +1-615-938-8594

E-Mail: akhileshraj91@gmail.com