Jeremiah Coholich

Robotics PhD Student

Georgia Institute of Technology

About Me

I am a Robotics PhD student at Georgia Tech in the Robotics Perception and Learning (RIPL) Lab, advised by Dr. Zsolt Kira. My current research thrust involves using computer vision techniques for robot learning.

In the summer of 2023 I was an intern at Honda Research Institute working on deep learning-based dexterous manipulation under Jinda Cui and Soshi Iba.

During my bachelor’s degree, I was an undergraduate research assistant in the Human Centered Robotics Lab where I had the pleasure of working with Dr. Luis Sentis and Dr. Gray Thomas. I spent my summers interning at SpaceX, Harmonic Bionics, and NASA JPL. Additionally, I spent a large amount of time co-founding and leading Longhorn Racing Electric.

I also enjoy hiking, running, and rock climbing.

Interests

Reinforcement Learning
Computer Vision
Robotics

Education

PhD in Robotics, 2025 (expected)
Georgia Institute of Technology
MS in Computer Science, 2022
Georgia Institute of Technology
BS in Mechanical Engineering, Highest Honors, 2019
The University of Texas at Austin

Publications

Soft Foot Sensor Design and Terrain Classification for Dynamic Legged Locomotion

Embedding sensors into a robotic foot and using them classify terrain with machine learning

Xiaofeng Guo, Bryan Blaise, Jennifer Molnar, Jeremiah Coholich, Shantanu Padte, Ye Zhao, Frank L. Hammond III

Soft Foot Sensor Design and Terrain Classification for Dynamic Legged Locomotion

Compliance Shaping for Control of Strength Amplification Exoskeletons with Elastic Cuffs

Designing hardware and a control algorithm for an adjustable-stiffness arm exoskeleton

Gray Cortright Thomas, Jeremiah M Coholich, Luis Sentis

Compliance Shaping for Control of Strength Amplification Exoskeletons with Elastic Cuffs

Evaluation of energy density measures and validation for powder bed fusion of polyamide

Characterizing the structual effects of laser energy density on 3D-printed nylon

David Bourell, Jeremiah Coholich, Antione Chalancon, Abhimanyu Bhat

Evaluation of energy density measures and validation for powder bed fusion of polyamide

Unpublished

Improving the Student-Teacher Approach for Semi-Supervised Semantic Segmentation

We propose and evaluate two variants to the mean teacher self-distillation technique (CS 8803 Learning with Limited Supervision final project)

Jeremiah Coholich, Harsh Maheshwari, Simar Kareer, Arun Ramachandran

Improving the Student-Teacher Approach for Semi-Supervised Semantic Segmentation

Learning High-Value Footstep Placements for Quadruped Robots

A method for choosing footstep placements over rough terrain which leverages a value function trained with reinforcement learning

Jeremiah M Coholich, Zsolt Kira

Learning High-Value Footstep Placements for Quadruped Robots

Generalizing Learned Policies to Unseen Environments using Meta Strategy Optimization

I benchmark Meta Strategy Optimization against Domain Randomization for sim2sim transfer. (CS 7643 Final Project)

Jeremiah Coholich

Generalizing Learned Policies to Unseen Environments using Meta Strategy Optimization

Sim2Real Transfer for Quadrupedal Locomotion

I review two sim2real transfer papers for robotics and then propose a new method using an “information bottleneck”. (Robotics PhD Qualifying Exam)

Jeremiah Coholich

Reinforcement Learning for Dynamical Systems

We evaluate the performance of reinforcement learning and other control algorithms on the Open AI Gym CartPole environment. (ECE 6254 final project)

Jeremiah Coholich, Cesar Santoyo, Angel Yam, Pratik Kunapuli

Reinforcement Learning for Dynamical Systems

Awards

National Defense Science and Engineering Graduate (NDSEG) Fellowship, 2020
NASA Space Technology Graduate Research Opportunities (NSTGRO) Fellowship, 2020 (declined)
Georgia Tech President’s Fellowship, 2019

Blog

A Bag of Tricks for Deep Reinforcement Learning

When I first started studying reinforcement learning (RL), I implemented Proximal Policy Optimization (PPO) from scratch using only the psuedocode on OpenAI’s website. It didn’t work and failed to obtain nearly any reward on most OpenAI Gym environments.

Jeremiah Coholich

May 31, 2022 16 min read

A Bag of Tricks for Deep Reinforcement Learning

I Completed the Google Foobar Challenge

The Google Foobar challenge is a programming challenge used by Google for recruiting. You must be invited to participate (article with more info). The challenge contains five levels of increasing difficulty, which require knowlege in the following areas:

Jeremiah Coholich

May 31, 2022 1 min read

I Completed the Google Foobar Challenge