Sitemap

We are preparing to use the direct method as in VO and formulate photometric error, I focused on photometric calibration when I learned and ported Direct Sparse Odometry Lite to our drone, which aims to provide localization in the wild and evaluate the importance of photometric calibration. I followed the manual photometric calibration paper from the TUM group and developed code to calculate the camera-response function and vignetting map. The calibration is proved to be useful for relative error metrics and prevents spikes in localization error.

I proved in details the convex relaxation portion of the paper Blind Deconvolution using Convex Program- ming and recreated its results. This project is completed with ECE273-convex optimization and greatly enhanced my understanding. My Instructor is Prof. Piya Pal.

I learned to use Isaac Gym and designed a network architecture that allows a high level network to select and combine appropriate low-level controllers. My network is proven to be beneficial. This is a course project instructed by Prof. Michael Yip.

A mini relational database implemented in C++ that supports sql-like commands with indexing and caching.

We designed PCB and control algorithm for this quadcopter which is ~14 cm in diagonal. This is a course project instructed by Prof. Steven Swanson.

Given stereo camera observation and IMU observations of a real self-driving dataset, I developed a sparse, feature based SLAM algorithm based on EKF and jointly estimation of agent pose and landmark positions. My Instructor is Prof. Nikolay Atanasov.

Given LIDAR and IMU observations of a real self-driving dataset, I developed a preliminary(no loop closure) SLAM algorithm based on occupancy map and particle filter. My Instructor is Prof. Nikolay Atanasov.

I want to make my powerhouse for my future robotics projects. The controller board senses the rotor's position from a magnetic encoder and modulates three phase voltage accordingly to achieve control goal. I have tuned the current loop to have a bandwidth of 2 kHz, and currently tuning the velocity and position loop. A command interface is also under construction.

This module is designed for localization need of my club Yonder Dynamics. I designed a compact PCB that holds IMU, pressure sensor, GPS, and magnetometer and communicates all information to ROS network through a ported rosserial package. I also learned to use ESKF to fuse all information.

I made a hand-held gimbal for stabilizing my phone when taking videos. The system cotains 3 BLDC motors and 2 IMUs, one at the handle and another attached to the end effector(phone holder). The difference between the IMU are used to calculate joint angles and guide the 3 phase voltage modulation.

This is our robot for the FRC 2019 competition. It can deliver hatch panels and fuels(big inflated balls) over the competition field. It re-uses the elevator to push its back wheels up the climbing zone and use the giant cylinders to climb. In particular, we achieved autonomous pose drive using a double PID loop.

Our robot for the 2018 FRC competition. It can deliver recycle boxes both to ground entrance and lift them up to around 2 meters and dump into a scale. For the game rule it can also hang itself up with a wrench. In this season I developed and installed a compact odometry module on the vehicle for our autonomous. We made it to world championship!

Robot for FRC 2017! Capable of launching balls into a cone that is ~2m above the ground. We used mecanum wheels and I added a gyroscope so that operators can control it in a headless mode.

Robot for 2016 FRC Stronghold. capable of catapulting ball to a target tower. Learned how to segment and track reflection markers using OpenCV.

Our FTC 2016 robot, capable of collecting balls and launch them to a basket. I added acceleration/deceleration sequence to the autonomous code to avoid slipping during the autonomous period.

Our FTC 2015 robot. The game requires robots to collect and deliver balls and cubes, then climb up a challenging ramp during the endgame

Our FTC 2014 robot. The game requires robots to collect balls on the ground and load them to a vertical tube. Unlike other teams that collect-raise-dump, we designed our robot to launch the balls up and guided directly into the tubes!

The first metal robot that I worked extensively! I am the programmer and learned to use encoders. Made several autonomous scripts and finally we got to National competition.

First ever metal robot that I worked with. Learned to use new hardwares such as servos. (then burned the servos the night before competition, sad).