top of page

Research Projects

Key words: Robotics, Optimal Control, Machine Learning, Motion Planning Algorithm

Receding Horizon Control Method for Moblie Robot based Signal Model Estimation

RA in Boston University

We formulated an optimization problem to minimize the estimation error for RSS (Received Signal Strength) model in a WSN (Wireless Sensor Network) by applying a MPC (Model Predictive Control) method. Then we applied a FIM(Fisher Information Matrix)-based RH (Receding Horizon) controller for agent motion planning and used MLE(Maximum Likelihood Estimator) for signal model estimation in real time. 

online-controler.png

Time Optimal Data Harvesting in Two Dimensions through Reinforcement Learning 

RA in Boston University

We proposed a multi-agent Mobile Robot data harvesting problem and developed a DRL (Deep Reinforcement Learning) solution to the time-optimal optimization. Then we applied a learning policy PPO (Proximal Policy Optimization) to control agent in a continuous action space, which exhibited Robustness to disturbances through the use of regularization for policy smoothing.

DH_RL_2023_2.png

Safety Guaranteed Optimal Control Policy for Data Harvesting using a CLF-CBF approach

RA in Boston University

We formulate a multi-agent data harvesting problem in 1.5-D space and generate a tracking trajectory by IPA gradient descent optimization. Then we develop a CLF-CBF based tracking controller to guarantee agent's safety in data harvesting mission. In the end, we solve the symmetric deadlock problem and provide simulation results in MATLAB.

deallock_case.png

Optical Flow Sensing Control Model using Deep Reinforcement Learning

RA in Boston University

We set up ROS and Gazebo environment for RL (Reinforcement Learning) training and robot control in Ubuntu 20.04. And then we build robot model with optical flow sensing camera and design RL training framework in Python. After that, we use Kalman filter to decrease the sensing noise, and the controller drive the robot running in the virtual environment successfully

OpticalFlow1.png

Design of a Gravity Driven Rolling Robot

Project in Columbia University

We design a 3-D robot model on Solid Works and finish the assembling work. Then we print the parts by 3-D printer, and install robot body with Raspberry Pi as the control center. In the end, the robot is driven by torque from its gravity, with 1.7kg wight and 3.1 d/c moving speed

RobotQ1.png

Robot System Design using Genetic Algorithm

Project in Columbia University

We use Genetic Algorithm to develop the robot structure designer in C++ Visual Studio. Then, we set up the simulation environment using OpenGL with physical interactive engine. After that, we successfully generate over 30 kinds of robots that have a speed over 0.5 diameter/cycle.

GeneticA1.png
bottom of page