Education
Ph.D. in Evolutionary Computation and Robotics
2018 – 2025
Indian Institute of Technology Kharagpur, West Bengal, India
Thesis: Multi- and Many-Objectives Optimization in Gait Generation of 25-DoF Humanoid Robot Using Genetic and Evolutionary Algorithms
Supervisors: Prof. D.K. Pratihar (IIT Kharagpur) and Prof. Kalyanmoy Deb (Michigan State University, USA)
M. Tech in Computer Integrated Manufacturing
2013 – 2015
National Institute of Technology Agartala, Tripura, India
Achievement: Gold Medalist (CGPA: 9.17)
B. Tech (Hons) in Mechanical and Automation
2007 – 2011
Maharshi Dayanand University, Rohtak, Haryana, India
Professional Experience
Assistant Professor
Jul 2025 – Present
BITS Pilani Digital, Birla Institute of Technology and Science, Bengaluru-560001, Karnataka, India
Digital Learning Division
Assistant Professor (On Contract)
Aug 2024 – May 2025
The LNM Institute of Information Technology, Jaipur
Mechanical-Mechatronics Department
Resource Person
Apr 2023 – Apr 2024
National Programme on Technology Enhanced Learning
Experimental Robotics - Conducted and Recorded Path and Gait Planning of Humanoid Robot
Assistant Professor
Jul 2015 – Dec 2017
Modern Institute of Technology and Research Centre, Alwar, Rajasthan
Research Interests
🤖 Machine Learning & AI
Applied Machine Learning
Deep Learning Techniques
Neural Networks
Pattern Recognition
Deep Learning Techniques
Neural Networks
Pattern Recognition
🔧 Optimization
Evolutionary Algorithms
Genetic Algorithms
Multi-objective Optimization
Swarm Intelligence
Genetic Algorithms
Multi-objective Optimization
Swarm Intelligence
🤖 Robotics
Humanoid Robot Gait
Path Planning
Control Systems
Biomechanics
Path Planning
Control Systems
Biomechanics
đź’» Soft Computing
Fuzzy Logic
Neural Networks
Genetic Programming
Hybrid Systems
Neural Networks
Genetic Programming
Hybrid Systems
Publications
Biomedical Robots and Devices in Healthcare: Opportunities and Challenges for Future Applications đź”—
Elsevier, December 2024
Biomedical Robots and Devices in Healthcare: Opportunities and Challenges for Future Applications explores recent advances and challenges involved in using these techniques in healthcare and biomedical engineering, offering insights and guidance to researchers, professionals, and graduate students interested in this area. The book covers key topics such as the current state-of-the-art in biomedical robotics and devices, the role of emerging technologies like artificial intelligence and machine learning, rehabilitation robotics, and the optimization techniques for optimal design and control. The book concludes by exploring the potential future developments and trends in the field of biomedical robotics and devices and their healthcare implications.-Provides a comprehensive overview of the current state-of-the-art in biomedical robotics and devices, including a discussion of the various types of devices and robots that are being developed and used in healthcare settings-Highlights the role of computational intelligence techniques such as artificial intelligence, machine learning, Fuzzy Logic, and evolutionary algorithms in the design, development, and the use of biomedical robots and devices, offering insights and guidance to professionals and students on these technologies-Explores the potential future developments and trends in the field of biomedical robotics and devices and their implications for healthcare professionals and patients, providing a valuable resource for those looking to stay up-to-date on advancements in the field
Analysis and Optimization of Gait Cycle of 25-dof NAO Robot using Particle Swarm Optimization and Genetic Algorithms đź”—
International Journal of Humanoid Robotics, World Scientific, Vol 21, No. 2, 2024
The gait cycle of 25-degree of freedom (DOF) humanoid robot, namely NAO robot, consists of single support phase (SSP) and double support phase (DSP). Both dynamic and stability analyses are carried out for this robot to determine its power consumption and dynamic stability margin, respectively. Constrained single-objective optimization problems are formulated for the SSP and DSP separately and solved using particle swarm optimization (PSO) and genetic algorithms (GA). A performance index, other than the fitness function, consisting of constraint values and maximum swing height, is also considered to compare PSO and GA-obtained optimal solutions. PSO is able to find the trajectories that offer higher swing height for nearly similar power consumption during SSP. A performance assessment of each algorithm based on the best fitness values in each generation across several runs is also carried out. These values are compared using the Wilcoxon rank-sum test, and PSO is found to be statistically better than GA. The optimal solutions from the simulations are tested using the Webots simulator to validate their efficacy on stability. Moreover, an investigation of the influence of gait parameters on power consumption during SSP and DSP reveals that the humanoid robot with a higher hip height, lower swing height, and slow pace consumes less power. The methodology developed in this is generic and can be easily extended to other robots.
Many-objective robust gait optimization for a 25-DOFs NAO robot using NSGA-III đź”—
Engineering Optimization, Taylor & Francis, 2025
Researchers increasingly employ evolutionary algorithms to tackle complex robotics problems owing to their ability to handle nonlinear dynamics. Robust solutions that remain stable despite variations in design parameters are essential for decision-makers. This study investigates Type I and Type II robustness approaches within constrained many-objective optimization (MaOO) frameworks, which are rarely explored in humanoid robotics. It focuses on optimizing the gait cycle of a 25-degrees-of-freedom (DOFs) NAO humanoid robot (NAO is an acronym for 'Nao d'Amour', also sometimes referred to as 'Now, Autonomous and Operative') during single- and double-support phases. The Non-dominated Sorting Genetic Algorithm III (NSGA-III) algorithm is employed to address four conflicting objectives, such as minimizing power consumption, maximizing stability, minimizing torque fluctuations, and reducing gait cycle time. A detailed comparative analysis highlights the superiority of Type II robustness, offering better-distributed and more convergent solutions in real-world scenarios with several constraints and variables. Furthermore, the study examines the influence of gait parameters on objectives, enhancing the understanding of humanoid robot dynamics and presenting a robust methodology for similar complex challenges.
Dynamic performance evaluation of evolutionary multi-objective optimization algorithms for gait cycle optimization of a 25-DOFs NAO humanoid robot đź”—
Swarm and Evolutionary Computation, Elsevier, 2025
Researchers are increasingly using optimization methods to achieve optimal dynamic performance of humanoid robots, often involving multiple conflicting objectives. Multi-objective optimization algorithms (MOAs) aim to find a Pareto front of optimal solutions, but selecting the best algorithm based on solution quality and computational efficiency remains challenging. This study comprehensively evaluates MOAs from different paradigms: swarm intelligence (CMOPSO), genetic algorithms (NSGA-II, DCNSGA-III), and decomposition-based approaches (CMOEA/D) for optimizing the gait cycle of a 25 DOF NAO humanoid robot during single support phase (SSP) and double support phase (DSP) scenarios. The algorithms' convergence, diversity, and constraint-handling capabilities are systematically analyzed in solving the gait generation problem. The bi-objective optimization simultaneously minimizes power consumption and maximizes dynamic stability subject to eight functional constraints with 12-13 decision parameters. Through performance evaluation using running inverted generational distance (IGD) and hypervolume (HV) metrics across eleven independent runs of each algorithm, NSGA-II emerges as the most suitable algorithm, demonstrating superior convergence and solution quality, while CMOPSO shows competitive performance with faster initial convergence. DCNSGA-III exhibits moderate performance with constraint-handling difficulties, and CMOEA/D demonstrates poor convergence characteristics requiring significantly more computational resources. Two distinct knee regions emerge during both SSP and DSP, representing optimal trade-off solutions, with a systematic framework provided for practitioners to select appropriate gait parameters based on operational priorities. The running IGD metric combined with HV validation demonstrates effectiveness in providing robust algorithmic insights, enabling practitioners to select suitable algorithms for similar complex real-world optimization problems.
Multi-and Many-Objective Optimization for Gait Generation of a 25-DOF Humanoid Robot Using Genetic and Evolutionary Algorithms đź”—
PhD Thesis, Indian Institute of Technology Kharagpur, 2025
Humanoid robots have attracted significant attention in recent years due to their potential to mimic human-like movements and perform various tasks in real-world environments. There is a growing interest among researchers to increasingly utilize Evolutionary Computation (EC) techniques to tackle complex robotics challenges, as these algorithms effectively manage the nonlinear dynamics inherent in humanoid robots. However, selecting the most suitable algorithm is challenging without a reference Pareto front (PF). Moreover, the entire region of the PF does not provide similar trade-offs. While most research has focused on optimizing only two to three objectives, bipedal locomotion inherently falls into the category of many-objective optimization with more than three competing objectives, also necessitating robust solutions that can withstand slight variations in design parameters to avoid unexpected results. This study presents a comprehensive investigation into the application of EC techniques to address these challenges, while optimizing the gait cycle of a 25 DOF NAO humanoid robot during both the single support phase (SSP) and the double support phase (DSP). Initially, constrained single-objective optimization problems for the SSP and DSP are solved using Particle Swarm Optimization (PSO) and Genetic Algorithms (GA). A comparison of the performances of PSO and GA is conducted over several runs by giving them equal opportunity to perform. PSO demonstrates superior performance compared to GA, attributed to its memory mechanism which enhances both local exploitation and global exploration capabilities. PSO outperforms GA in finding trajectories that offer higher swing height for nearly similar power consumption during SSP. The optimal solutions are validated using the Webots simulator. Next, the study considers two objectives simultaneously, viz., minimizing power consumption and maximizing stability. These are solved using the elitist non-dominated sorting genetic algorithm-II (NSGA-II) and multi-objective PSO (MOPSO). These EMO algorithms are compared using the Dominance Move indicator, formulated as a Mixed Integer Programming problem. The running performance metric is utilized to study the convergence and diversity behavior of MOPSO and NSGA-II. The results indicate that NSGA-II is a better choice for solving complex problems involving many constraints and variables. Two knee-finding methodologies, angle- and utility-based methods, are applied within NSGA-II to focus on the knee regions (KRs) of the PF to reduce the burden on decision-makers in selecting the most preferred solutions. These methods replace the crowding distance operator in NSGA-II to focus on KRs. The utility-based method is found to be more effective in identifying well-distributed solutions near KRs compared to the angle-based approach. Finally, constrained many-objective optimization problems are formulated for each gait scenario. The NSGA-III algorithm is employed to solve four conflicting objectives simultaneously, viz., minimizing power consumption, maximizing stability, minimizing torque fluctuations, and minimizing gait cycle time. The analysis reveals that Type-II robustness is superior, providing better-distributed and more convergent solutions, especially in complex real-world problems with several constraints and variables. The study on the influence of gait parameters on objective functions reveals that the higher hip height, lower swing height, and slower pace lead to lower power consumption. The faster speeds enhance stability at the cost of increased power consumption. The arm movements, particularly elbow roll, significantly enhance the stability without substantially increasing power consumption, especially during SSP. The velocities between SSP and DSP are crucial for smooth gait transitions. These findings highlight the importance of carefully balancing gait parameters to achieve optimal performance in different scenarios, whether prioritizing stability on challenging terrains or energy conservation in resource-limited environments. This study contributes significantly to the advancement of EC techniques in humanoid robotics, offering a comprehensive analysis of single-objective, multi-objective, and many-objective optimization approaches for gait cycle optimization. The findings not only deepen the understanding of humanoid robot dynamics but also introduce robust methodologies that can be extended to other complex real-world problems.
Effect of B2O3 containing fluxes on the microstructure and mechanical properties in submerged arc welded mild steel plates đź”—
IOP Conference Series: Materials Science and Engineering, IOP Publishing, 2016
This paper represents a study on the effect of B2O3 additions in fluxes on the microstructure and mechanical properties of the weld metal formed during Submerged Arc Welding of Mild Steel plates. Five fluxes with about 2.5, 5, 7.5, 10 and 12.5% B2O3 were used with a low carbon electrode. Welding process parameters were kept constant for all the conditions. The microstructure of weld metal for each flux consisted mainly of acicular ferrite, polygonal ferrite, grain boundary ferrites and equiaxed pearlite. It was noted that the Vicker's hardness value was a function of boron content and shows a mixed trend. Impact Energy and Tensile Strength were increased with the increase in boron content in welds this can be attributed to relation with the higher acicular ferrite percentage. However an optimum level of toughness and tensile strength was available with 7.5% and 5% of B2O3 respectively. A qualitative comparison has also be done with fresh flux by means of full metallography and mechanically.
A Knee-Based Multi-objective Optimization for Gait Cycle of 25-DOF NAO Humanoid Robot đź”—
Lecture Notes in Networks and Systems, vol 994, Springer, 2024
A multi-objective optimization problem finds multiple optimal solutions represented on a Pareto front (PF), for conflicting objectives. Focusing on the "knee" region (KR) of the PF is preferred to targeting the entire PF since there is a significant degradation in one objective for a minor gain in another outside the KR. This paper applies two knee-finding methodologies—angle- and utility-based methods within the elitist non-dominated sorting genetic algorithm (NSGA-II), to address a multi-objective optimization problem of a 25-DOF NAO humanoid robot's gait cycle. The objectives are minimizing power consumption and maximizing dynamic balance margin. The single support phase exhibits a single KR, whereas the double support phase shows two KRs. This research demonstrates a knee-based multi-objective optimization algorithm to reduce the burden on decision-makers in selecting the most preferred solutions. It compares two knee-finding techniques and provides insights into a practical robotics problem for different gait cycle phases.
A Comparative Study of Pareto Front of Optimal Solution Set for NAO Robot's Gait Optimization Using the Dominance Move Indicator Based on Mixed Integer Programming đź”—
Lecture Notes in Mechanical Engineering, Springer, 2025
Evolutionary multi-objective optimization (EMO) algorithms generate solution sets representing trade-offs between conflicting objectives. The comparison between two EMO algorithm performances is challenging for real-world problems lacking reference Pareto fronts (PFs). Most performance indicators require a reference PF or point to compare the Pareto-optimal solutions. This study uses Dominance Move (DoM), formulated as a Mixed Integer Programming (MIP), as it compares sets without any reference PF or points and avoids information loss. The GD+, IGD+, HV, Epsilon, and MIP-DoM indicators have been used to compare solution sets from two popular EMO algorithms—NSGA-II and MOPSO. EMO algorithms are used to minimize power consumption and maximize stability for a 25-DOF humanoid robot gait optimization problem. The results show that NSGA-II outperforms MOPSO on this highly constrained problem. The MIP-DoM exhibits the strongest correlation with the IGD+ indicator, whereas weaker correlations are seen for the Hypervolume and Epsilon indicators. The EMO performance has also been tracked over generations using IGD+, which provides additional insight into algorithm dynamics. The proposed techniques could be extended to other real-world optimization problems.
Multi-objective optimization of rotational magnetorheological abrasive flow finishing process đź”—
Nanofinishing of Materials for Advanced Industrial Applications, CRC Press / Taylor & Francis, 2023
The rotational magnetorheological abrasive flow finishing (R-MRAFF) technique achieves uniform, nano-level mirror finishes with high material removal (MR) rates, distinguishing it from other nano-finishing methods by combining rotational motion and magnetorheological (MaR) abrasive particles during the nano-finishing process. Regression analysis is conducted to assess the influence of input process parameters, namely, extrusion pressure (P), finishing cycles (N), rotational speed of the magnet (S), and abrasive mesh size (M) on the responses like percentage improvement in surface roughness (%ΔRa) and amount of material removed (MR). The obtained regression equations for %ΔRa and MR are then used to formulate a multi-objective optimization problem, which is solved by an elitist non-dominated sorting genetic algorithm-II (NSGA-II). The final results revealed a trade-off between these two objectives. The higher P, N, and S levels effectively generated a trade-off for the better surface finish (SF) and a good MR. However, the lower levels of M are adequate for both the responses. The study's findings, particularly the identified optimal parameters' combinations, offer valuable insights for maximizing the potential of R-MRAFF, enabling the attainment of desired SF and material removal characteristics in a range of applications. This study can be extended to other complex manufacturing processes with multiple parameters and responses.
Comparative evaluation of deep learning techniques for multistage Alzheimer's prediction from magnetic resonance images đź”—
Biomedical Robots and Devices in Healthcare: Opportunities and Challenges for Future Applications, Elsevier, 2025
This study applies deep learning (DL) techniques to predict different stages of Alzheimer's disease (AD) from magnetic resonance imaging (MRI) scans. AD is a neurodegenerative disorder that causes progressive dementia. Early diagnosis can help delay cognitive deterioration. The dataset consists of 6400 MRI images labeled as nondemented, very mild demented, mild demented, and moderate demented. Augmentation techniques are used to create 12,800 training images to balance the classes. Three DL models are tested on multiclass AD prediction: convolutional neural networks (CNN), ResNet-50, and VGG-16. The models are trained for 100 epochs on categorical cross-entropy loss with 80:20 train-test splits. VGG-16 achieves the best testing accuracy of 89.92%, followed by CNN (85.12%) and ResNet-50 (71.01%). VGG-16 also shows the highest precision, recall, and F1 scores for all four classes. It performs well in the mildly demented class, which is difficult for the other models. The results confirm that complex deep CNNs can detect subtle imaging features related to early AD changes. Transfer learning with pretrained models yields better results than training simple CNNs from scratch. The study offers insights into the best DL approaches for computer-aided diagnosis of AD using MRI data.
Soft robotics and computational intelligence: Transformative technologies reshaping biomedical engineering đź”—
Biomedical Robots and Devices in Healthcare: Opportunities and Challenges for Future Applications, Elsevier, 2025
The introductory chapter begins by examining how emerging technologies are transforming biomedical engineering and healthcare. It highlights the importance of computational intelligence, which includes machine learning and deep learning, for diagnosing diseases, personalizing treatments, and optimizing resources. The chapter then introduces soft robotics—a fast-growing field that offers potential benefits for surgery, rehabilitation, prosthetics, and assistive devices due to its gentle interaction with tissues. It describes the manufacturing methods such as elastomer curing and additive manufacturing, and the applications such as surgical systems, exosuits, prosthetics, and artificial organs. It also discusses the main advantages of soft robotics, such as biocompatibility, maneuverability, and fatigue resistance, as well as the current challenges, such as miniaturization, power supply, and modeling complexity. The chapter ends by summarizing the detailed analyses of various innovations in the following chapters—covering topics such as multibistatic medical sensing, mathematical modeling, sleep analysis using wearables, ankle-foot orthosis design, machine learning for prognosis, disease prediction, automation-based advancements, brain activation topography, mental health frameworks, and wearable sensors with on-body electronics. The chapter illuminates the fusion of technologies like soft robotics, artificial intelligence, and wearables in shaping the future of healthcare.
Advancing ankle–foot orthosis design through biomechanics, robotics, and additive manufacturing: A review 🔗
Biomedical Robots and Devices in Healthcare: Opportunities and Challenges for Future Applications, Elsevier, 2025
Orthotic devices play a pivotal role in lower limb rehabilitation by providing support, stabilization, and movement assistance. However, their design must be optimized to achieve the desired functional outcomes and enhanced mobility. This comprehensive chapter reviews the biomechanics, design principles, manufacturing methods, and emerging technologies for advancing ankle–foot orthosis (AFO) functionality to improve lower limb rehabilitation. It analyzes ankle–foot biomechanics during gait to inform AFO design. Various types of AFO and their applications for spinal cord injury, stroke, and other conditions are discussed. Additive manufacturing techniques like 3D printing enable customized, patient-matched AFO fabrication. Integrating sensors, actuators, and control systems into powered AFOs promotes movement and neuroplasticity. However, technical challenges remain around device weight, cost, user acceptance, power needs, and safety. Future directions involve leveraging AI for personalized control, remote monitoring, and telemetry-based expert oversight. Overall, this research-backed review elucidates optimizing AFO design through integrating biomechanics, robotics, and advanced manufacturing to enhance mobility and quality of life.
A Research Perspective on Ankle–Foot Prosthetics Designs for Transtibial Amputees 🔗
Mechanical Engineering in Biomedical Applications, Scrivener Publishing, Wiley, 2024
Amputees may regain their lost limbs and continue daily activities using prosthetic ankle–foot devices, and researchers have presented various prosthetic foot designs that could improve functionality during the last several decades. However, a vast number of individuals are denied access to this technology, since most of the developed prosthetic devices are exorbitantly bulky in design and expensive. Thus, engaging amputees in different activities, such as walking, jogging, dancing, cycling, golfing, swimming, and other sports with a single design, becomes impossible. The current study aims to synthesize existing information on ankle– foot prosthetics based on demographics, engineering designs, and biomechanics underlying the biological foot to identify significant challenges and limitations of existing work to lead the future development of low-cost ankle prostheses. Based on the findings of the previous studies, it can be concluded that future research should focus on a variety of factors, including but not limited to low cost, domestic accessibility, robustness, moisture and corrosion resistance, susceptibility to easy fabrication, lightweight, suitable aesthetic, and psychosocial acceptability.
Effect of Boron Trioxide Enriched Fluxes on the Microstructure and Mechanical Properties in Submerged Arc Welded Mild Steel Plates đź”—
Advanced Aspects of Engineering Research Vol. 1, Book Pi, 2021
Among the several arc welding methods, the submerged arc welding is the preferred method for welding thick sections in the industry because of its several advantages which include high production rates, good weld quality, ease of automation and minimum operator skill requirement. This paper represents a study on the effect of B2O3 additions in fluxes on the microstructure and mechanical properties of the weld metal formed during Submerged Arc Welding of Mild Steel plates. Five fluxes with about 2.5, 5, 7.5, 10 and 12.5% B2O3 were used with a low carbon electrode. Welding process parameters were kept constant for all the conditions. The microstructure of weld metal for each flux consisted mainly of acicular ferrite, polygonal ferrite, grain boundary ferrites and equiaxed pearlite. It was noted that Vicker's hardness value was a function of boron content and showed a mixed trend. Impact Energy and Tensile Strength were increased with the increase in boron content in welds this can be attributed to relation with the higher acicular ferrite percentage. However, an optimum level of toughness and tensile strength was available with 7.5% and 5% of B2O3 respectively. A qualitative comparison has also been made with fresh flux through full metallography and mechanically.
Designing Driver Drowsiness Detection Systems: Challenges and Solutions
International Journal of Commercial Vehicles, SAE International, 2025 [Under Review]
Conducted comprehensive review of state-of-the-art driver drowsiness detection systems and highlighted main challenges in existing methods. Focused on how to develop an Adaptive Driver Drowsiness Alert System (ADDAS) that uses responsible AI techniques to provide personalized and secure drowsiness prediction. Suggested solutions for feature discrepancy, privacy concerns, model robustness, and explainability challenges using transduction transfer learning and multimodal data integration.
Novel Machine Vision Method for Unwrapping Images of Pipelines
23rd ISME International Conference on Recent Advances in Mechanical Engineering (ICRAME-2025), December 17-19, 2025 [Under Review]
Developed a novel machine vision framework for unwrapping pipeline images by imposing and exploiting scene constraints for efficient inspection. Integrated Canny edge detection with geometric transformation models to achieve accurate surface reconstruction with 93.2% surface area recovery after unwrapping. Addressed fundamental challenges in computer vision-based pipeline inspection by providing simplified yet effective solution for pixel scale calculation using system constraints.
Achieving Smooth Gait Transitions in Bipedal Locomotion through Double Support Phase Integration in the 3D Linear Inverted Pendulum Model
23rd ISME International Conference on Recent Advances in Mechanical Engineering (ICRAME-2025), December 17-19, 2025 [Under Review]
This research focuses on achieving smooth gait transitions in bipedal locomotion through the integration of double support phase in the 3D Linear Inverted Pendulum Model. The study addresses the challenges of maintaining stability and continuity during the transition between single and double support phases in humanoid robot walking.
Nonlinear Inverse Design Optimization for Parameter Assessment of a Porous Rectangular Fin
28th National and 6th International ISHMT-ASTFE Heat and Mass Transfer Conference (IHMTC-2025), December 9-12, 2025 [Under Review]
This study presents a nonlinear inverse design optimization approach for parameter assessment of a porous rectangular fin. The research employs advanced optimization techniques to determine optimal design parameters that enhance heat transfer performance in porous fin structures.
Inverse Thermodynamic Optimization of Input Parameters for a Biomass-Powered Bigeneration Cycle
28th National and 6th International ISHMT-ASTFE Heat and Mass Transfer Conference (IHMTC-2025), December 9-12, 2025 [Under Review]
This research focuses on inverse thermodynamic optimization of input parameters for a biomass-powered bigeneration cycle. The study employs optimization algorithms to determine optimal operating conditions that maximize energy efficiency and sustainability in biomass-based power generation systems.
Projects
Academic Research Projects
Optimal Gait Generation of Bipedal Locomotion using Evolutionary Computation
2018 – 2025
Ph.D. Thesis
Focused on developing robust and stable gaits for bipedal locomotion in humanoid robots using evolutionary computation techniques. Formulated single/multi/many-objective optimization problems for single and double support gait phases and solved them using genetic and evolutionary algorithms. Introduced novel methods to compare algorithms, prioritized knee-region solutions on the Pareto Front for decision-making, and ensured robustness in locomotion.
Effect of Enriched Fluxes on the Microstructure and Mechanical Properties
2014 – 2015
M. Tech Dissertation
The effects of TiO2 and B2O3 on the microstructure and mechanical properties of high-strength and mild steel during the Submerged Arc Welding process were investigated. Metallurgical microstructure and mechanical properties data were collected, and the Grey Relational Coefficient was applied to determine the best process parameters for optimizing the desired mechanical properties.
A Micro-Controller Based Robotic Arm for Autonomous Material Handling
2010 – 2011
B. Tech Final Year Project
An attempt was made to create a material handling device employing a robotic arm that operates autonomously upon receiving a signal from the micro-controller. It can be implemented in any automated material handling system where destination is already known.
Machine Learning Based Projects
Designing Driver Drowsiness Detection Systems
2024 – 2025
Research Output: Communicated with International Journal of Commercial Vehicles, SAE International
- Conducted comprehensive review of state-of-the-art driver drowsiness detection systems and highlighted main challenges in existing methods
- Focused on how to develop an Adaptive Driver Drowsiness Alert System (ADDAS) that uses responsible AI techniques to provide personalized and secure drowsiness prediction
- Suggested solutions for feature discrepancy, privacy concerns, model robustness, and explainability challenges using transduction transfer learning and multimodal data integration
Novel Machine Vision Method for Unwrapping Images of Pipelines
2024 – 2025
Research Output: Communicated with ISME 2025 Conference
- Developed a novel machine vision framework for unwrapping pipeline images by imposing and exploiting scene constraints for efficient inspection
- Integrated Canny edge detection with geometric transformation models to achieve accurate surface reconstruction with 93.2% surface area recovery after unwrapping
- Addressed fundamental challenges in computer vision-based pipeline inspection by providing simplified yet effective solution for pixel scale calculation using system constraints
Deep Learning Techniques Comparison for Multi-Stage Alzheimer's Prediction
2023 – 2024
Research Output: Published as Book Chapter in Biomedical Robots and Devices in Healthcare, Elsevier, 2025
- Applied deep learning techniques (CNN, VGG-16, and ResNet-50) to predict different stages of Alzheimer's disease from magnetic resonance imaging (MRI) scans
- Implemented data augmentation techniques to create balanced dataset of 12,800 training images from original 6,400 MRI images labeled across four dementia classes
- Achieved 90% testing accuracy with VGG-16 model, demonstrating superior performance in multi-class classification with highest precision, recall, and F1-scores for all dementia stages
Projects as Senior Research Fellow (SRF/JRF)
Failure Analysis of Reformer PIGTAILS
Nov 2022 – May 2023
Consultancy Project | Funding: â‚ą9.8 Lakhs
Funding Agency: MATIX Fertilisers and Chemicals Ltd
Objective: Investigated and found remedies to prevent the initiation of cracks in Austenitic Stainless Steel (SS304H) material due to high-temperature fluid exposure in Matix Fertilisers & Chemicals Limited
Supervision: Prof. D.K. Pratihar (Principal Investigator) and Prof. Debalay Chakrabarti (co-PI), IIT Kharagpur, West Bengal, India
Status: Completed
High-Speed Walking Gait Control of a Life-Size Humanoid Robot
March 2019 – Sept 2021
Sponsored Project | Funding: â‚ą1.7 Lakhs
Funding Agency: Shashtri Institute, Delhi
Objective: The robot was investigated and studied for defence and humanitarian aid applications
Supervision: Prof. D.K. Pratihar (Principal Investigator), IIT Kharagpur, West Bengal, India and Prof. Alejandro RamĂrez-Serrano, University of Calgary, Alberta, Canada
Status: Completed
Guest Lectures
Theory to Code: Hands-on ML, Genetic Algorithms, and Fuzzy Logic with Python
March 2024
Geethanjali College of Engineering and Technology, Hyderabad
Delivered lectures in the 5-day Faculty Development Program organized by the Department of CSE (Cyber Security)
Current Trends in Mechanical Engineering: Case Studies From Industries and Academia to Promote Innovation, Design Thinking and Startups
May 2024
Shri Vishnu Engineering College for Women, Vishnupur, Bhimavaram
Delivered a lecture in the 5-day Faculty Development Program organized by the Department of Mechanical Engineering
Quantum Machine Learning
May 2024
Mercedes-Benz India
Delivered an online lecture on Quantum Machine Learning
Genetic Algorithm-based Approach to Improve Humanoid Robot Stability on Different Terrains
November 2019
IIT Kharagpur
Delivered a lecture as part of the "AICTE-QIP Short-Term Course on Robotics"
Technical Skills
đź’» Programming
Python
MATLAB
Wolfram
R
MATLAB
Wolfram
R
🎨 Design Software
AutoCAD
SolidWorks
CAD Modeling
3D Design
SolidWorks
CAD Modeling
3D Design
đź“š Academic
LaTeX Typesetting
Research Methodology
Technical Writing
Data Analysis
Research Methodology
Technical Writing
Data Analysis
🗣️ Languages
Hindi (Native)
English (Fluent)
Technical Communication
Academic Presentation
English (Fluent)
Technical Communication
Academic Presentation
Contact Information
Let's Connect!
📧 Email
pushpendra050@gmail.com