I am a first-year Robotics Masters student at the University of Michigan (UM) advised by Prof. Honglak Lee. My research focuses on developing data-driven algorithms to enable efficient and generalizable robot learning for long-horizon compositional tasks. I am interested in Deep Learning and its applications towards robotic perception, control, and decision-making. I aim to deploy these algorithms onto real-world robotic systems such as personal assistants and medical robots.
I completed my Bachelors in Computer Science with a minor in Mathematics at UM. This past summer, I had the wonderful opportunity to work with Prof. Changliu Liu at Carnegie Mellon University's Robotics Institute to develop a safe control architecture for self-driving in dynamic environments. In previous summers, I've completed industry internships at Invisible.ai, Google Ads, Luminar and Socratic.
Feel free to say hi: aliangdw at umich dot edu
Carnegie Mellon University RI
May 2020 - Present
AI Research Intern
May 2020 - Aug 2020
University of Michigan
Computer Science and Math
Sept 2017 - May 2020
Google Ads Quality
Software Engineering Intern
May 2019 - Aug 2019
AI Engineering Intern
May 2018 - August 2018
Sept 2017 - May 2018
Software Engineering Intern
May 2017 - Aug 2017
|Oct 2020||Preprint out on arXiv: Reinforcement Learning for Sparse-Reward Object-Interaction Tasks|
|Sept 2020||Started as a Robotics Masters student at the University of Michigan.|
|May 2020||Joined the Intelligent Controls Lab as a Visiting Scholar with Prof. Changliu Liu!|
|May 2020||Graduated with a Bachelors degree in Computer Science and minor in Mathematics!|
Optimal Control and Reinforcement Learning
Weiye Zhao*, Anthony Liang*, Rui Chen, Changliu Liu
Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in First-person Simulated 3D Environments
Wilka Carvalho, Anthony Liang, Kimin Lee, Sungryull Sohn, Honglak Lee, Richard L. Lewis, Satinder Singh
NeurIPS Deep Reinforcement Learning 2020 Workshop
ROMA: A Relational Object Modeling Agent for Sample-Efficient Reinforcement Learning
Wilka Carvalho, Anthony Liang, Kimin Lee, Sungryull Sohn, Honglak Lee, Richard L Lewis, and Satinder Singh
ICML Object-Oriented Learning 2020 Workshop (Oral Presentation)
Drift-Aware Predictive Coding
Domain shifting dynamics is informative for making predictions in a gradually shifting environment. In this project, we propose to directly learn the environment shift dynamics through representation learning. For each domain, our proposed approach learns a Drift-Aware Predictive (DAP) coding, which embeds information helpful for predicting in the next domain, and modulates a base learner (e.g. a classifier in a classification problem) to adapt to the shifted new domain.
Natural Language Processing Class Competition
In our final project, we designed and implemented models to benchmark three different natural language tasks: CommonsenseQA (CSQA), Conversational Entailment, and Everyday Actions in Text (EAT). We experiment with different preprocessing techniques, model architectures, and hyperparameters. Our team placed 1/25 in the class competition for the EAT task and 2/25 in both CSQA and Conversational Entailment.
MRover Robotic Arm
Full stack library for operating an autonomous robotic arm on a space rover. Includes modules for perception, kinematics, motion control, planning, self/world obstacle collision, etc. Developed a web interface using KinEval to visual robot arm and instruct arm to navigate to waypoints.
Invariant EKF for Robot Localization using IMU and GPS
Matlab implementations of EKF and IEKF for localizing a robot traveling around North Campus done for a graduate level Mobile Robotics course. We use IMU data to predict motion and GPS data to correct the robot motion.
Social Media Based Movie Recommender System
A web-based movie recommendation system based on a person's social media profiles. We implemented large-scale web scrapers for retrieving information from social media profiles and movie datasets. Movies and users are represented using a vector space model and each movie is ranked based on a cosine similarity score.
Implemented a real-time barcode detection system based on RBox, a variant of the object bounding boxes that accounts for rotation of objects. Model is used in industrial settings for scanning barcodes that are often in different lighting and orientations.