Guining Pertin

M.S Robotics, KAIST

Senior Engineer, Enphase Energy

B.Tech ECE, IIT Guwahati

Hello There!

Welcome to my portfolio site,

I recently graduated as a Masters student in The Robotics Program at KAIST, South Korea where I also worked as a Graduate Research Assistant at Control Lab under Professor Dong Eui Chang. I completed my B.Tech in Electronics and Communication Engineering from IIT Guwahati, India in 2021, and was a Senior Engineer at Enphase Energy, India before I joined graduate studies.

My interests lie primarily in nonlinear systems, reinforcement learning, data-driven control and developing simulation models for the same. Other than research and engineering, I love teaching, reading, music, badminton and video games (playing/building).

To search through all my projects, please use this big ol' search bar:

Featured Projects

Master's Thesis: Foundations

This project explores and designs different approaches to involve physical knowledge (or model-based designs) into purely learning-based paradigms to improve sample efficiency and robustness for both drones and manipulators.

Master's Thesis: Differentiable Simulations

This section of my Master's thesis considers a relatively new approach for accelerated policy learning by directly backpropagating gradients through the simulation steps, using the new Nvidia Newton physics engine.

SleeperAgent: Dreaming from a third perspective

This project considers a new approach to utilize wrist-only (or ego) images from manipulators to predict the 3rd person views through novel view synthesis to improve performance along with Dreamer-V3 world-model based RL.

Bachelor's Thesis: Multi-agent multi-objective optimization using deep Q-learning

My BTech thesis on exploring multi-agent reinforcement learning by combining classical path planning and coordinated learning for the agents through a novel state representation with branched DDQN algorithm.

MoPAT Mk. V: Motion Planning Algorithms Testbed

A ROS2 based open-source, low-cost robot with 2D motion capture platform. The original version used ArUco tags for an inter-hostel competition I hosted, but the final version uses RGB LED patterns to localize and track multiple robots.

Project Zenith: Multi-UAV search and rescue

An event hosted by the Defence Research and Development Organisation for autonomous search-and-rescue missions using multi-UAV swarm for accurate detection and GPS localization of targets. Our team won SILVER.