Projects

A novel method for autonomous flight of MAVs in dynamic uncertain environments without making specific assumptions on the obstacles.
Published at RA-L

Detection and trajectory estimation algorithm fusing event stream and depth estimation
Algorithms are integrated in a MAV system
Published to IROS 2021

This is a summer research training project for 2nd-year undergraduates guided by PhDs at XJTU.
The goal is to improve the performance of their ICCV by changing network structure and applying self-supervised learning tricks.
Skills I learned from this project includes pytorch, training deep learning model skills and software engineering skills

1. Reproduced the research paper "Learning Monocular Dense Depth from Events" from scratch
2. Trained the reproduced model on DSEC dataset
3. Improved the performance by introducing SSIM loss
This is the course project of Seminar Computer Vision by Deep Learning at TU Delft.

1. Implemented PRM planner and KD tree
2. Implemented A-star method for graph search
3. Implemented classic minimum-snap trajectory optimization (both optimization and closed-form)
4. Modified above optimization methods to confine hard-constraints.
This is the course project of Planning & Decision Making at TU Delft.

1. Modelled the problem by its piecewise affine (PWA) approximation
2. Transformed PWA model to Mixed Logical Dynamical (MLD) system
3. Designed an MPC controller for the MLD model, and rewrote it to MILP formulation
4. Applied both explicit and implicit MPC to find the optimal control policy
This is the course assignment of Modeling and Control of Hybrid Systems at TU Delft.

Implement solutions to the multi-aircraft coordination formulation problem based on dual decomposition, Nesterov method, incremental subgradient method and consensus ADMM approach. Investigated convergence of the optimization process under different parameter settings.
This is the course assignment of Networked and Distributed Control Systems at TU Delft.

We employ a nonlinear model predictive control approach to guide a drone in catching a thrown ball.
This is the course project of Model Predictive Control at TU Delft.

1. The motion planning part is based on MoveIt and teb_local_planner
2. The object perception is achieved by detecting AprilTag markers on the objects
3. The planning domain definition language is applied for high-level reasoning and task planning
4. The demonstration in Gazebo simulation is finished as the course project of Knowledge Representation and Symbolic Reasoning at TU Delft.
5. The real-world demonstration is achieved as parts of the Mulitdisciplinary Project

Implement Karto-SLAM on a turtlebot robot with RPLIDAR-A2. Robots can also be controlled by a xbox one controller via bluetooth.
This is part of a 4th-year undergraduate course at XJTU.
Software implementation is based on https://github.com/mywisdomfly/Clean-robot-turtlebot3

We develope a remote-controlled car based on Xlinx EGO-1 FPGA board as a microcontroller. The car is equipped with L298N stepper motor drivers and 4 two-phase four-wire stepper motors. The FPGA board is utilized for bluetooth communication, instruction encoding and stepper motor control. User can send instructions with desired direction and speed via Bluetooth to control the car. The proposed design demonstrates the capabilities of FPGA-based systems for implementing complex control functions in robotic applications.
This is the project of a 3rd-year undergraduate course at XJTU.

This training project involves building a 450mm-frame drone from scratch, covering mechanical assembly, electrical welding, computer environment setup, software deployment, and real-world tests.
It requires a thorough understanding of drone mechanics, electrical engineering, and software programming. The construction process involves selecting appropriate parts, soldering and wiring the components, and ensuring proper balance for stable flight.

In this project, we implement a monocular depth estimation network using Tensorflow 2 based on published papers.