About me

Hi, I'm a CS Ph.D. student at the University of Pennsylvania advised by Dinesh Jayaraman. My goal is to develop robots can learn and act by themselves. I received my BS/MS in CS from the University of Southern California, where I worked with Joseph J. Lim. I also developed software for nonprofits at Code The Change, and interned at Tesla and Intel.


I aim to develop robot learning algorithms that can curate their own data through autonomous exploration, leverage easy to obtain priors, and learn efficiently through model-based reinforcement learning.


  • ๐Ÿ”ญ Exploration: The key bottleneck to robot learning is data. If we want a robot to learn many different behaviors, a diverse dataset is needed for training it. Therefore, exploration, or the process of collecting the data, is of paramount importance.
  • ๐Ÿ—ƒ๏ธ Convenient Priors: Giving prior knowledge can be very powerful but the most helpful priors are typically the most expensive. I am interested in designing priors that are easy to specify and methods that can maximally exploit them.
  • ๐Ÿ”ฎ World Models: Endowing robots with a world model, or a model that predicts the dynamics of the world, is very useful. A robot can generate data by predicting outcomes, or leverage the model to make informed decisions not achievable by a model-free method.

Research

image for IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks
Planning Goals for Exploration ๐Ÿ”ญ๐Ÿ”ฎ
Edward S. Hu, Richard Chang, Oleh Rybkin, Dinesh Jayaraman

In submission to ICLR'23.
CoRL Workshop on Learning, Perception, and Abstraction for Long-Horizon Planning, 2022 (Oral)

image for IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks
Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning ๐Ÿ—ƒ๏ธ
Kun Huang, Edward S. Hu, Dinesh Jayaraman

Conference on Robot Learning, 2022 (Oral, 6.5% acceptance rate), Best Paper Nominee

image for IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks
Transferable Visual Control Policies Through Robot-Awareness ๐Ÿ—ƒ๏ธ๐Ÿ”ฎ
Edward S. Hu, Kun Huang, Oleh Rybkin, Dinesh Jayaraman

International Conference on Learning Representations (ICLR), 2022
ICLR Workshop on Generalizable Policy Learning, 2022 (Oral, 5% acceptance rate)

image for IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks
IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks

International Conference on Robotics and Automation (ICRA), 2021

image for To Follow or not to Follow: Selective Imitation Learning from Observations
To Follow or not to Follow: Selective Imitation Learning from Observations ๐Ÿ—ƒ๏ธ

Conference on Robot Learning (CoRL), 2019

Composing Complex Skills by Learning Transition Policies

International Conference on Learning Representations (ICLR), 2019

Fun Projects

Optical Illusion in Python
Make black and white images seem vibrantly colored with a sparsely colored grid overlay.

GPT-2 Slackbot
Enables slack users to converse with OpenAIโ€™s GPT-2 language model

Medmind
Open-source medication management software.

Mentorship

Current:
Past:
  • Kun Huang, Prev. ROBO MS @ UPenn. Now ML Eng. @ Cruise