Panlong Gu

Postgraduate Student of Control Engineering

Education

MEng in Control Engineering, 2019.09 - now

Shandong University
BSc in Measurement and control technology and instrument, 2015.09 - 2019.06

Qufu Normal University

Interests

Robotics
Visual SLAM (Visual Simultaneous Localization and Mapping)
3D Target Detection and Tracking
Fusion of SLAM and target detection

Biography

Hi, I’m Panlong Gu (顾潘龙), a Master candidate of the School of Control Science and Engineering at Shandong University (山东大学控制科学与工程学院), Jinan, Shandong, China.

In the past postgraduate life, as a main research member, I participated in the research projects of the laboratory. In the national key research and development plan "service robot cloud service platform" project, I designed a trajectory instance segmentation method to extract all independent dynamic trajectories which are distributed in the observation space of the robot. Besides, in this project, I also researched and wrote the semantic lidar mapping algorithm and the 3D semantic mapping algorithm based on scene graph enhancement;

In the Shandong Province Major Science and Technology Innovation Project "The key technology research and industry of cloud-based home service robot practicalization". I conducted a study of the Cartographer and the RGBD fusion mapping method. I have completed the conversion of relevant invention patents and papers for the related achievements of the above projects.

At the same time, I have also participated in the writing of the laboratory project plane and have certain experience in project declaration. In addition, during my master’s and undergraduate years, I actively participated in various domestic and provincial competitions in related disciplines. I won two national first prizes, two national second prizes, two national third prizes, and provincial champions, 15 prizes in provincial and below competitions

In the current research, I will focus on the mutual integration of SLAM and 3D target detection, and plan to introduce SLAM as an independent module into 3D target detection to improve the single reasoning ability of the recognition algorithm and the generalization ability of the model. Moreover, it can complete the camera SLAM process, and output the feature point map and camera trajectory with 3D Bounding Box annotations. At the same time, by relying on the feature point matching process in the SLAM algorithm, the tracking of the three-dimensional object can be completed with the lowest computational cost, and its pose information can be continuously optimized.

Skills

Robotics

Ubuntu, ROS, image processing methods, PID Control, Sensor Filter, Navigation, etc.

SLAM

Mapping, Location, Semantic SLAM, SLAM enhanced target detection, etc

Deep Learning

2D target detection, 2D instance segmentation, 3D target detection, 3D instance segmentation, etc.

Programming

C++, Python, C, Matlab, pytorch, tensorflow, etc.

Embedden System

K60, STM32, CAN, UART, IIC, SPI, etc.

Design

Altium Designer for hardware, SolidWorks for machine

Experience

Docking Representatives

National Key R&D Program Funded Project Robot Group, Shandong University & Inspur.

Nov. 2017 – Now Jinan. Shandong, China.

Responsibilities include:

Cooperate with Inspur to complete the national key R & D plan;
3D Semantic Mapping and Object Relation Detection;
Navigation with 3D Semantic Map and Scene Graph.

Leader

Robot Perception Group, Cloud Intelligent Robotics Laboratory, Shandong University.

Jan. 2020 – Now Jinan. Shandong, China.

Responsibilities include:

SLAM based 3D Object Tracking;
Multi-motion Visual Odometry;
Graph CNN based Dynamic Trajectory Segmentation;
SLAM based Stereo 3D Object Detection.

Member

Robot Perception Group, Cloud Intelligent Robotics Laboratory, Shandong Univeresity.

Sept. 2018 – Now. Jinan. Shandong, China.

Responsibilities include:

Cartogrpher based Large Scene Mapping;
A 3D Reconstruction Method Using Multisensor Fusion in Large-Scale Indoor Scenes.

Vice Chairman of the Student Union

Qufu Normal University (QFNU)

Sept. 2017 – Jun. 2018. Rizhao, Shandong, China.

Responsibilities include:

As the representative of the college, organize the selection of related competitions within the discipline;
To coordinate the work of science and Technology Department, network department, work department and other departments in the student union;
Organization the related activities within the College;
Related work docking of colleges and Universities.

Secretary of the Youth League Branch

Measurement and Control Class of QFNU

Sept. 2015 – Jun. 2018. Rizhao, Shandong, China.

Responsibilities include:

Class affairs management;
The construction of the Communist Youth League in the class.

Award

The 16th China post-graduate mathematical contest in modeling

National First Prize · Dec. 2019

1: We use the conversion relationship of the relevant coordinate systems in the camera imaging process to establish a camera calibration model based on the vanishing point and the vanishing line, and obtain the internal and external parameters of the camera based on the prior scale information in the scene. And established the relative depth solution model of the camera to recover the relevant scale information in the picture;

2: We combine the known scale information in the image and the SFM algorithm to estimate the relevant scale information in the video.

The 15th China Graduate Electronic Design Contest

National Third prize · Aug. 2020

Cloud-based robotic arm cloth stacking based on behavioral cloning in a human-machine collaborative environment.

We use Apriltag technology to fix the workbench at the set position in the pixel coordinate system, and use image segmentation technology and deep reinforcement learning to complete the stacking of cloth and the cloning of human behavior

National Undergraduate Electronic Design Contest

National second prize · Aug 2017

Our team designed a ball and board control system with a camera. We use the cascade PID algorithm to set the speed and position closed loop. Through the return of IMU data and the control of the putter motor, a closed-loop system is formed to realize the control performance of the ball movement.

Shandong Province Intelligent Control Competition

Champion · Aug 2017

We recognize the lane line through the visual scheme, and use this to control the differential Ackerman model for racing competitions.

China Engineering Robotics and International Open (ROBOWORK)

National third place (National First Prize) · May 2018

We control the quad-rotor drone through a visual solution to complete the drone's automatic tracking and take-off and landing process.

China Robotics Competition

National Secound Prize · May 2017

We recognize different types of obstacles (high platforms, steps) through visual solutions, and use this to control the quadruped robot to cross obstacles and navigate

Tianchi Big Data Contest Protein Structure Prediction Contest

No. 4/241 teams worldwide · Mar 2020

We use the physical and chemical characteristics of the protein to combine with the tree model to predict the secondary structure based on the primary structure of the protein.

In addition, there are more than 15 provincial competition awards

Projects

Graph CNN based Multi-motion Visual Odometry

This project uses the graph convolutional neural network to transform the removal of dynamic feature points into a two-class anomaly detection problem with unbalanced samples, and uses the back-end spatio-temporal dual-track VO to estimate the movement of the instance and the camera.

3D Semantic Scene Map

In this project, we captured more details in the environment and output the semantic relationship between objects, allowing the robot to understand the concept of "a notebook on the table".

Semantic Lidar Mapping

The project uses 2D object detection and gmapping to solve the 3D bounding box of the object, and generate a lidar semantic map.

Multi-Sensor 3D Reconstruction

This project completes 3D Reconstruction by inserting RGBD point clouds into the Submap, which generated by Cartographer.

Outdoor cruise robot

We designed a set of automatic mapping and cruise system in the outdoor distribution project, using 16-line lidar and differential GPS to complete the robot's outdoor navigation and mapping.

3D Semantic Scene Map

In this project, we captured more details in the environment and output the semantic relationship between objects, allowing the robot to understand the concept of "a notebook on the table".

Semantic Lidar Mapping

The project uses 2D object detection and gmapping to solve the 3D bounding box of the object, and generate a lidar semantic map.

Multi-Sensor 3D Reconstruction

This project completes 3D Reconstruction by inserting RGBD point clouds into the Submap, which generated by Cartographer.

Outdoor cruise robot

We designed a set of automatic mapping and cruise system in the outdoor distribution project, using 16-line lidar and differential GPS to complete the robot's outdoor navigation and mapping.

Graph CNN based Multi-motion Visual Odometry

3D Semantic Scene Map

In this project, we captured more details in the environment and output the semantic relationship between objects, allowing the robot to understand the concept of "a notebook on the table".

Semantic Lidar Mapping

The project uses 2D object detection and gmapping to solve the 3D bounding box of the object, and generate a lidar semantic map.