Welcome to my portfolio
Nature answers all our questions as poems written in the form of Mathematics. All we need to do is Interpret
About me
I'm a forward-thinking Engineer who loves to be challenged by interesting problems. Being a Computer Science and Physics enthusiast, I believe in interdisciplinary research.
Currently, I'm pursuing Master's in Robotics engineering at WPI and an active researcher at Controls and Reinforcement Learning Lab.
The curriculum, being project-based, has helped me develop a pragmatic outlook and the skills required to realize scientific theories in the real world. I've worked on projects like Decentralized reinforcement learning for multiagents, Optimal Control of Handle robot, Active Planning in Stochastic Systems, Skill Learning, and High-Level Motion Planning, Learning Obstacle Avoidance using Deep Reinforcement Learning to name a few. Additionally, the coursework encouraged me to actually implement different filtering techniques, mapping and motion planning algorithms, reinforcement learning and control algorithms for a deeper understanding of their characteristics and implications.
Being a Reinforcement Learning summer intern at Mathworks not only gave me theoretical exposure and implementation experience but also motivated me to develop an optimistic outlook towards the field. Work experience at a Pharmaceutical Packaging company and a software firm have helped me develop skills like writing clean and reusable code, working effectively as a part of a cross-functional team, use of different collaboration tools and team/project management
I am willing to contribute to the field of Motion Planning and Control Systems with an emphasis on Learning.
Skills
My Projects
Decentralized Multi-agent Reinforcement Learning
This project aims at studying the current centralized and decentralized learning algorithms and develop a quick decentralized algorithm still capable of learning performance comparable to centralized algorithms. The simulation testbed comprises of a navigation task assigned to two turtlebots and was developed by me.
Active Planning in Stochastic Systems
Human Learning is a result of processing new data and updating belief depending on its fidelity. This inspires us to develop a similar framework for robot agents to proactively plan in real time given the perception cues which works by striking a balance between exploration & exploitation through intrinsic motivation.
Quadrotor Learning Obstacle Avoidance
The project aimed to demonstrate a successful application of Deep Q Learning to train a quadrotor to learn to avoid obstacles using just visual cues.
High Level Motion Planning
Robots can perform much more complicated tasks if the information can be abstracted appropriately to plan at a higher level.
This project aims to develop a framework to learn the PDDL representation of the task which would enable human-like planning for complex tasks.
Laser Surgery using ABB IRB120
Free hand surgeries by surgeons lack the degree of precision, especially where the contour of ablation is complicated.
This project aims at developing a framework to enable Laser Surgery using robot manipulators to enable faster, safer and high dexterity ablations.
Underactuated balancing of Handle robot
This project considers the case of failed knee and hip joint actuators the Handle robot and tries to balance it upright using just motor actuation at feet. The robot is modeled as a triple pendulum mounted on a moving cart and an LQR control law is implemented to balance the system upright with actuation being solely the cart.
