Kartik Paigwar

I am a Machine Learning Engineer at Tenstorrent Inc. working on a new generation AI compute. Previously, I have worked at Verdant Robotics, Luminosity Labs and IISc Bangalore where I had a chance to work with Dr. Gabe Sibley, Tyler Smith, Prof. Shishir Kolathaya and Prof. Shalabh Bhatnagar. My area of interests includes reinforcement learning, deep learning and robotics.

I finished my Master's degree from ASU with a specialization in Robotics and AI. During my graduation, I did research on agile sim-to-real locomotion in collaboration with Improbable AI Lab, MIT advised by Prof. Pulkit Agrawal. I had a fun time working on Mini-Cheetah robot, making it run faster and jump over large gaps.

I received my bachelor’s degree in Computer Science from Visvesvaraya National Institute of Technology (VNIT), India in September 2019. I was an undergraduate summer research fellow at AIRLab, Politecnico di Milano, Italy advised by Prof. Andrea Bonarini and Dr. Davide Tateo. I worked on inverse reinforcement learning and dimensionality reduction for a search and rescue robotics task. During my bachelors, I was one of the core-coordinators of IvLabs, the robotics lab at VNIT where I spent most of the days and nights learning and teaching robotics.

Email  /  CV  /  Google Scholar  /  Github  /  Linkedin  /  Twitter

• February 2023: Starting as a ML Engineer at Tenstorrent Inc., Boston, USA.
• April 2022: Our paper on "Rapid Locomotion via Reinforcement Learning" got accepted in RSS 2022, NYC, USA.
• September 2021: Our paper on "Learning to Jump from Pixels" got accepted in CoRL 2021, London, UK.
• October 2020: Our paper on "Robust Quadrupedal Locomotion on Sloped Terrains: A Linear Policy Approach" got accepted in CoRL 2020, MIT, USA.
• August 2020: Had a wonderful discussion on Sim2Real with Prof. Rupam Mahmood (University of Alberta) and Jessica Hamrick (DeepMind) as a part of 1:1 sessions at CIFAR DLRL Summer School.
• June 2020: Our paper on "Learning Stable Manoeuvres for Quadruped Robots from Expert Demonstrations" got accepted in RO-MAN 2020, Naples, Italy.
• June 2020: Accepted to 2020's cohort of Deep Learning Reinforcement Learning Summer School hosted by CIFAR, Mila, Alberta Machine Intelligence Institute (Amii), and Vector Institute. I'm among the 300 applicants selected across the world.

I work on efficient ways an AI agent, robots or a silicon can be trained and learn important interactions with the physical world.

High-speed running and spinning on diverse terrains with a single neural network.

A hierarchical control framework for dynamic vision-aware locomotion.

What is the minimum possible control framework that can be deployed to realize stable locomotion behaviors on slopped terrains in medium-size low-cost quadruped robots?

Generating stable foot trajectories for Omni-directional quadruped motion and learning smooth transitions between these trajectories using expert demonstration.

With a view toward fast deployment of learned locomotion gaits in low-cost hardware, we generate a library of walking trajectories, namely, forward trot, backward trot, side-step, and turn in our custom-built quadruped robot, Stoch 2, using reinforcement learning.

We present a deep learning based approach for stair detection, statistical filtering on images for the estimation of stair alignment, and novel mechanical design for an autonomous stair climbing robot. The primary objective is to solve the problem of indoor locomotion over staircases

An omni-directional visual imaging system is constructed using a paraboloid reflector and a monocular camera as a cost effective on-board solution for mobile robot navigation.

A Deep RL framework for autonomous skills acquisition in which an agent learns from expert’s gameplays to exhibit a repertoire of skills in an adaptive game environment.

Worked on an inverse reinforcement learning problem to find a reward function which could explain the strategies incorporated for robot teleoperation during search and rescue missions.

Networks can be trained to fuse mutliple images of a same scene with different focal settings and capture a fully focused image with a minimal specification smart phone camera.

Rubik's Cube Solver(RCS) is a complete program that can solve any scrambled 3X3X3 cube in less than 22 moves. It uses kociemba algorithm for finding the most optimum solution of a scrambled cube.

Faculty

Collaborators

Past Mentees

Yes! You have guessed right. This guy has made a nice webpage.