Hi there! I'm Pang (Jeff) Liu, an AI & Robotic enthusiastic.
I focus on AI applications in solving real world problems, especially in Robotics field.
I have a dream that one day, automated robots will handle all repetitive tasks,
making the work much easier and allowing people have more time to relax!
Color Calibration Card
MLOps
ML
ETL
ETV
MTL
MRD
RandomForest
GBT
small neural network
Multi-Functional Cleaning Robot
This project has launched both in software and hardware fronts.
I build a turtlebot3 robot, with functions
I am developing a smart cleaning robot with frontier-based exploration, path planning, route following and voice control.
[->GitHub]
Map Processing
Path Planning
GUI Control
Voice Control
AI Agent on Sepsis Task
An AI Agent framework, desgined for seamless extension into domain-specific fields like Finance and Healthcare.
See whole report in Research Notes section.
[->GitHub]
AI Agent
AI Infra
RAG
Tool-augmented Reasoning
Prompt Engineering
Value-based Reinforcement Learning Methods on Breakout Game
Explore designing and implementing value-based methods on breakout game.
[->GitHub]
Q-learning
DQN
RL
CUDA
Smart Glove which can control Robot Arm
An AI-powered and raspberrypi-pico based smart glove that can control the robot arm.
[->GitHub]
Embedded Development
Gesture Recognition
Policy-based Reinforcement Learning Methods on Walker2d task
Explore designing and implementing policy-based methods on walker2d task.
[->GitHub]
MuJoCo
Policy-Gradient Control
Actor-Critic
PPO
RL
IoT Platform (Collaborated Project)
This project builds a scalable IoT platform for real-time monitoring and control of diverse devices. Using MQTT and a Raspberry Pi hub, it supports remote setup, data display, and rule-based automation.
With contributions from
Aby Iberkleid,
James Petullo, and
Shubham Kaushik.
MQTT
ROS
React
Edge AI
Following Line with Color-based Centroid Detection
This program can allow robot following a line and avoid obstacles.
The following logic is based on color and centroid detection.
[->GitHub]
ROS1
OpenCV
PID Control
Escape the Maze with Left Wall Following
This is my first project and designed at Brandeis Robotics lab.
This project is to escape the maze with left wall following algorithm.
[->GitHub]
ROS1
Linux
LiDAR
Solve Blackjack - Monte Carlo ES
The game blackjack was simulated 2 million times using the Monte Carlo ES method and converged to the optimal solution.
[->GitHub]
Reinforcement Learning
Monte Carlo ES
Fiducials Detection
Implemented a real-time ArUco fiducial marker detection system using computer vision techniques.
[->GitHub]
Computer Vision
ArUco Markers
Fiducials
Intelligent NPC Demo
This is the prototype or demo of project "Work With Farmer Louise".
The NPC know nothing at first, all skills need to be learned from player.
[->GitHub]
OpenAI API
Prompt Engineering
Pygame
A* Route Planner
Use open source map to implement A* route plan.
The start position is Tufts University,
and the end point is in a park near Mystic River.
No Demo, See Source Codes:
[->GitHub]
OSM
C++
A*
Ichiban Lottery (Commercial Project)
The Tianyuan Ichiban Lottery is a WeChat Mini Program offering anime figure lottery products via a digital platform system.
Deployed in real market and attracted real users, the program generated
over $15,000 in revenue.
-
This report compare different optimizers including SGD, AdaGrad, RMSProp, and Adam, to analyze the impact of adaptive gradient and momentum to the neural network's training.
(2025-10-09)
-
This report applies Q-learning and DQN to Atari game Breakout (Gymnasium v5) using PyTorch, serving as an educational resource.
(2025-05-06)
-
This report discusses how to use an AI Agent to finish a medical sepsis task and discussed the design of a simple AI agent.
(2025-05-02)
-
This study uses ML tools to develop and validate a robust color calibration model that achieves precise color reproduction in a real-time MLOps system.
(2025-03-14)
-
This is a tiny-research that extracting colors in HSV format and mark the centroid to achieve single color line following.
(2024-11-15)
-
This is a tiny-research on maze escape solution.
(2024-10-27)
DS&A Roadmap
This roadmap includes foundational data structures and algorithms, give you a clear study route that can learn and prepare for technique interviews.
Beyond this, the advanced algorithms are also attached and contain Codeforces or AtCoder links for those who attempt to participate in advanced algorithm competitions.
AI Notebook (Chinese Only)
Currently only support Chinese. This is a notebook containing AI foundational knowledges, including Artificial Intelligence, Machine Learning, Reinforcement Learning, and Deep Learning.
My first language is Chinese, so normally taking Chinese notes firstly. In view of the large amount and content of AI notes, I currently not consider translating them into English.
