深蓝学院

shenlanxueyuan.com

Motion Planning for

Mobile Robots

主讲人 Fei Gao

Ph.D.in Robotics
Hong Kong University of Science and Technology
Tenured Associate Professor,Zhejiang University
Introduction

2
About this Course
• This course is about:
• Academism (old school) planning pipeline
• Path finding algorithm
• Trajectory generation/optimization
• Real-time software implementation

• This course is NOT about:

• Dynamics Modelling
• Robot design
• End-to-end navigation
• Learning-based method
• Mathematical/theoretical proof

• This course may also cover:

• Autonomy for mobile robots
• Paper recommendations
• Cutting-edge research direction/ethics

3
Basic Expectation

• Discipline:
• Every researcher has his own taste/style. Since you choose this course, please follow my style.
• Q&A only at appointed office time.
• Finish your homework by yourself.

• Project:
• Basic algorithm validation (MATLAB)
• Sophisticated engineering implementation (ROS/C++)

每位研究者都有自己的风格和品味，如果你们选择了这门课，那请你们在这门课的范围内，尽量
follow我们的风格。我相信，如果大家跟随老师的风格，无论如何，大家最后都会收获很多很多。

我简单说一下我的风格： 我的风格一定是实用第一 ，任何算法如果没有很好的实用价值，那我就认为

它在实际的机器人系统上没有太大实际意义。我教给大家的算法，希望不要仅停留在Matlab的数值验
证层面，更要在C++或者ROS里面跑起来，在嵌入式系统中跑起来。这就是我的风格！

4
Teaching Team
Instructor 1:Fei Gao Instructor 2:Hongkai Ye
Ph.D.in Robotics
HongKong University of Science and Technology D.Eng.in Robotics
Zhejiang University
Assistant Professor,Zhejiang University

Instructor 3:Zhepei Wang Instructor 4:Jialin Ji

Ph.D.in Robotics D.Eng.Candidate in Robotics
Zhejiang University Zhejiang University

Instructor 5:Xin Zhou Instructor 6:Qianhao Wang

D.Eng.Candidate in Robotics Ph.D.Candidate in Robotics
Zhejiang University
Zhejiang University

5
Workload

• Expected Student Background:

• Linear algebra
• Probability
• MATLAB programming skills
• C++ programming skills (VERY IMPORTANT)
• Linux
• Love robots

• Workload:
• Attend lectures
• Lots of project work
• Have fun with robots

6
 What is autonomous robot

Definition: an autonomous robot is a robot that performs behaviors or tasks with a high degree
of autonomy (without external influence).

Subjects: Computer Science, Automation, Mechanism, Electronics …

• Perception-Planning-Control action loop

Localization

Sensing Planning Control

Mapping

7
 What is autonomous robot

estimation
l-single robot
★ ★ ★

mapping planning control

Depth Sensor
S。 Onboard computer
Autopilot

coordinate navigation -swarm

Data share

tmin tmax
 Vision-based Autonomous Drone
Planning
Perception Core
Path Path Trajectory Traj; Trajectory Cmd.
Searching Optimization Server
Map

Sensor Perception Execution

Image
Camera Dense Mapping Controller
Pose
IMU State Estimator Actuator

Real-time monocular dense mapping on aerial robots using visual-inertial fusion,ICRA 2017
Autonomous robot:applications

3D laser scanner

IMU
PS antenna

ereo camera
ear
cameral
2D laser
scanner
canner

Encoder

10
Aerial Robots

11
Dam Inspection

Shen,et al,2014 12
Dam Inspection

Ozaslan,et al,2014
13
Cellular Tower Inspection

14
Precision Farming

2X
Shen,et al,2014 15
Delivery

30SESSION
16
Ground Robots

17
Search and Rescue

18
Housekeeping

iRobot
19
Hills Driving

20
Urban Driving

⑪TierM AXZ
Intelligent Vehicle アイサフ于JDS- 株式会社
21
Crowd Driving

22
What is autonomous robot
· Estimation
Low latency
High accuracy &consistency

· Perception
3D sensing &dense perception
Map fusion &integration for planning

Filving Mors Aggressvely

· Planning
Complex &unknown environments
Safety &dynamical feasibility
Limited sensing &computation
Gap pose:
Roll =45°
Pitch =0°

Control
Yaw= 0
·
Aggressive maneuvers
Smooth trajectory tracking
What is motion planning

• Basic requirements
Safety: collision avoidance
Smoothness: energy saving, comfort
Kinodynamic feasibility: executable, controllable

• Old-school pipeline
Front-end path finding
Search for an initial safe path

Low dimensional
Discrete space

Back-end trajectory generation

Search for an executable trajectory
High dimensional
Continuous space

24
How to do robotics research

• Find a problem
A problem only in your imagination is not a problem at all.
A robotician must be an engineer first.
Hot topic is just hot.
Be honest.

• Solve a problem
Don’t intentionally making things complicated.
Simple but effective solution is always preferable.
Simulation tells nothing, show everyone real robots.
Solve real problems.
Solve it 100%, or not.

25
How to do motion planning

• Overall knowledge of planning

Choose suitable methods for different scenarios.
Design customized strategy.

• Dirty hands
Don’twait, don’t just read papers. Do it yourself.
A lot of coding work.
A lot of field work.

• Know the whole system well

Take care every component in your robot.

26