Junchao Li
Development
IA, United States
Skills
Machine Learning (ML)
About
Junchao Li's skills align with IT R&D Professionals (Information and Communication Technology). Junchao also has skills associated with Electrical Engineers, Designers, and Technicians (Engineering). Junchao Li appears to be a low-to-mid level candidate, with 5 years of experience.
View more
Work Experience
Research Assistant of Robotics
Controls & Autonomous Systems Lab
June 2019 - September 2020
- The University of Iowa Managed the lab's daily affairs and coordinated key experimental operations. Maintained and programmed KUKA KMR platforms; two basic demonstration examples: Video Link.
Research Assistant
Multiscale Computational Science and Engineering Laboratory
March 2020 - Present
- The University of Iowa Developed model-based DRL algorithms for robotic motion planning in partially observable environment for complex tasks. Employed POMDP for environment modeling and LTL for complex task specifications. Incorporating CNN in Deep Q-Network (DQN) architecture for feature extraction of belief state, aiding in Q value prediction. Proposed model-free DRL algorithms for solving robotic motion planning problems in partially observable environment with labeling uncertainty for complex tasks. Pioneered the use of Probabilistically-Labeled POMDP (PL-POMDP) to model the partially observable environment with dynamic events. Designed a model-free DQN architecture that integrates RNN (LSTM) to process the sequential inputs of observations and task recognitions, for Q value prediction. Created a safe and reliable autonomous agent for ethical decision-making through DRL. LTL was employed to specify safety restrictions and 'hard' ethical constraints, and reward design to handle 'soft' ethical constraints to guide the AI decision-making subject to the ethical norms. Developed an optimized traffic light control agent on Simulation of Urban Mobility (SUMO) using DRL. An MDP model was constructed from the 'position', 'velocity', 'waiting time', and 'driving directions' of the vehicles in SUMO. Deployed a CNN-enhanced DQN to predict Q values, optimizing traffic flow and control.
Education
The University of Iowa
Bachelor of Science in Mechanical Engineering
September 2013 - September 2017
