Shuai Wang

Ph. D. Candidate of Robotics.

University of Science and Technology of China

Biography

I am now a Ph. D. student at the School of Information Science and Technology, University of Science and Technology of China, China, working with Prof. Jiahu Qin on Autonomous Navigation of Mobile Robots.

My research interest lies in the intersection of Perception and control in robotics, such as the End-to-End Driving System. I’m also interested in Multi-Modal Data Fusion, Autonomous Navigation of Mobile Robots, Multi-Robot Formation Control, etc. Currently, I am focusing on applying deep learning technology to the issue of decision-making of autonomous driving vehicles.

Interests

End-to-End Driving System
3D Perception and control
Multi-Sensor Fusion
Computer Vision
Autonomous Vehicles
Multi-Robot System

Education

Ph.D. in Control. Sci. Eng., 2022
University of Science and Technology of China
M.Sc. in Control. Sci. Eng., 2018
University of Science and Technology of China
B.Sc. in Automation, 2016
Northeast Forestry University

Publications

Quickly discover relevant content by filtering publications.

FlowDriveNet: An End-to-End Network for Learning Driving Policies from Image Optical Flow and LiDAR Point Flow

Introduction Methodology Experiments

Shuai Wang, Jiahu Qin, Menglin Li and Yaonan Wang

PDF Code Video

FlowDriveNet: An End-to-End Network for Learning Driving Policies from Image Optical Flow and LiDAR Point Flow

Automated Visual Surveillance Using Multiple Cooperative Mobile Robots

Introduction Methodology Experiments

Shuai Wang, Yuhang Chen, Jiahu Qin and Tao Yang

Code Video

Automated Visual Surveillance Using Multiple Cooperative Mobile Robots

Spatio-Temporal Ultrasonic Dataset: Learning Driving from Spatial and Temporal Ultrasonic Cues

Introduction Methodology Experiments

Shuai Wang, Jiahu Qin, and Zhanpeng Zhang

PDF Code Video

Spatio-Temporal Ultrasonic Dataset: Learning Driving from Spatial and Temporal Ultrasonic Cues

Circular Formation Algorithms for Multiple Nonholonomic Mobile Robots: An Optimization-Based Approach

Introduction Experiments

Jiahu Qin, Shuai Wang, Yu Kang and Qingchen Liu

PDF Code Video

Circular Formation Algorithms for Multiple Nonholonomic Mobile Robots: An Optimization-Based Approach

A Novel Variable-Gain Rectilinear or Circular Formation Algorithm for Unicycle Type Vehicles

Shuai Wang, Jiahu Qin, Qingchen Liu and Yu Kang

PDF

A Novel Variable-Gain Rectilinear or Circular Formation Algorithm for Unicycle Type Vehicles

See all publications

1. Introduction The simplest form of Imitation Learning is Behavior Cloning (BC), which focuses on learning the expert’s policy using supervised learning. Behavioral cloning is a method by which human sub-cognitive skills can be captured and reproduced in a computer program.

Code Video

End-to-end Driving via Imitation Learning

Competitions

Huawei CodeCraft 2021

First prize of Shanghai-Hefei Competition Area

Mar 2021 – Apr 2021 Team Member

The competition focuses on the problem of Resource Planning and Scheduling on the cloud.
In the Preliminary and Semi-Finals, the optimal deployment strategy of each deployment request of the virtual machine must be given in real time to reduce the server cost as much as possible. We design a comprehensive evaluation metric by considering various indicators such as energy consumption, matching degree of demand types, and the size of remaining resources. Finally, a real-time deployment strategy was achieved, and the energy cost was further reduced by resource arrangement.
In the Finals, demand pricing and opponents’ game are considered, and it is expected to maximize the order profit. We use machine learning to regress the opponent’s pricing strategy for achieving dynamic game, and combine our well-defined breakeven price achieving the order profit maximization.

DeeCamp 2020 Auto Driving Competition

Second Place in Self-driving Track, First Place in Lidar Group

Jun 2020 – Aug 2020 Team Leader

The goal of this competition is to realize 3D Object Detection of various common objects in the urban environments while ensuring inferencing speed based on the high-precision 128-Line LiDAR dataset provided by Momenta.
The overall solution mainly includes Data Preprocessing, Point Cloud Data Enhancement, Detector and Tracker building and Result Visualization.
I am mainly responsible for the Detector Design and Model Tuning based on SECOND model with Det3D framework and the Visualization of detection results.

UISEE Auto 2031 Self-Driving Challenge

First Place in the Preliminary, Second Place in the Semi-Final and Final

Sep 2019 – Dec 2019 Team Leader

This competition requires competitors to build a self-driving system from scratch based on a function limited self-driving simulator provided by the UISEE company.
The ultimate goal is to realize the autonomous driving of vehicles in simulation with the fastest speed in the unknown test track.
I am mainly responsible for the End-to-End Driving System design and implementation, including remote control, data collection and processing, data enhancement, model design and tuning, model deployment and vehicle lateral and longitudinal control. Finally, the maximum driving speed of 170km/h is achieved by the Optical-Flow based driving model and adaptive vehicle longitudinal control.