Kartik Garg

I am an MS Robotics student at Georgia Tech. I work with Prof. Sehoon Ha on RL for humanoids. 

Over the summer, I worked as an AI Resident at 1X, where I taught NEO cool stuff with RL :). I completed my undergrad at the Indian Institute of Technology (BHU) Varanasi,  pursuing my bachelor's in Mechanical Engineering.

Previously, I have:

•  Worked at IISc Bangalore at the Stochastic Robotics lab headed by Prof. Shishir Kolathaya on making a robot dog see and walk.

• Collaborated with Dr. Sourav Garg from AIML, University of Adelaide, to develop viewpoint invariant visual place recognition systems.

• Interned at the Flight Robotics and Perception Group, University of Stuttgart as a DAAD WISE Scholar, German-Indo exchange program. I developed a deep-RL-based drone docking system under the supervision of Prof. Aamir Ahmad and Pascal Goldschmid. 

In my free time I quiz and design cool posters. I also practice kendo (read training to be a samurai :)

I am always happy to learn and open to network.    

By applying unsupervised reinforcement learning, we trained agile locomotion behaviors without relying on a curriculum or reward engineering.

Webpage

Developing viewpoint invariant visual recognition models by utilizing the zero-shot capabilities of large vision models like SAM, and DINOv2. 

Webpage Code Paper  

Developed control algorithm for powered prosthetic foot. This project’s aim was to make a computationally effective classification algorithm to detect human gait phases. A major limitation of existing powered foot prosthetics is that they use high power computational hardware which is costly and makes it undesirable for the majority of the Indian demographic. I formulated a fuzzy logic-based algorithm to perform real-time classification. The algorithm produced exemplary real-time results, which led to a journal publication in Springer Nature, Computer Science Journal.

Paper  

Developed a unified diffusion-based policy for morphology adaptive multi-robot multi-skill locomotion. 

Video

Developed visual locomotion policies using deep reinforcement learning to enable locomotion on challenging terrains.