Projects

Projects

[PUBLICATION PREPRINT] Reduced Variance Deep Reinforcement Learning with Linear Temporal Logic

Labels: Deep Reinforcement Learning, Robot Motion Planning, Linear Temporal Logic(LTL)

Applied deep reinforcement learning techniques to solve LTL-based robot motion planning problems. And developed reduced variance algorithms to stabilized the output from the Neural Network.

[UNDER REVIEW] Deep Imitative Reinforcement Learning for Temporal Logic Robot Motion Planning with Noisy Semantic Observations.

Labels: Deep Reinforcement Learning, Imitation Learning, Robot Motion Planning, Model Checking

Designed planning experts that can guide the Q-learning agent toward convergence.

Mapping the Ocean Floor — Classifying underwater sonar data using Convolutional Neural Network (CNN)

Labels: Deep Neural Network, Acoustics, Sonar, MFCC

Applied CNN on simulated sonar data to identify underwater hills and valleys.

                                         

Product Automaton — A correct implementation of taking product between Non-deterministic Büchi Automaton and Finite Transition System

Labels: Motion Planning, Linear Temporal Logic, Python, MATLAB

After figuring out some existing MATLAB package cannot correctly converting Linear Temporal Logic (LTL) to Büchi Automaton (ltl2ba), if negation exists in LTL, and hence cannot product a correct Product Automaton. I decided to create my own package.

Impact of Underlying Voting Distribution in Societal Tradeoff Rules

Labels: Societal Tradeoffs, Economics, Artificial Intelligence

This project explored a way optimizing the allocation of societal resources among voters and evaluated the reliability of existing approaches.

Distributed Robotics

Labels: Distributed Systems, Artificial Intelligence, Algorithms

Explored how systems of mobile robots interacting collaboratively. Experimented with cases such as how to cooperate on common tasks, and how to coordinate their motion through the world to reach individual goals.

Amazon Picking Challenge

Labels: Robotics, Gripper Design, System Integration

As part of Team Duke, I worked on designing a robot gripper that can grab and keep items with various shape and texture accurately and firmly and integrated it into a robotic system which can identify, pick up and collect items autonomously.

Analyzing Self-Designed Universal Beam by Finite Element Method

Labels: Universal Beam, Finite Element Method, Euler-Bernoulli Beam Theory

In this project, I designed a lightweight I-beam that can be applied on overhead crane for performance improvement. The yield strength and yield moment has been analyzed through finite element method and the result shows that the crane is well-supported by the newly designed beam in simulated working environment.

Miscellaneous

Brief Review of Poisson Image Editing

Labels: Image Editing, Numerical Analysis, Iterative Method

Summarized how to solve optimization problems in poisson editing by numerical iterative methods. (Click for the writeup)

Financial Case Study on Montessori Charter Schools

Labels: Social Innovation, Entrepreneurship, Financial Report

Charter schools play essential roles in primary and secondary education. However, a large part of closed Charter Schools are triggered by financial deficiencies. In this project I conducted case study on successful and unsuccessful Montessori Charter Schools and analyze the feasibility of constructing a successful Charter School in Raleigh-Durham area. (Click for the Writeup)