AUTONOMOUS ROBOTICS

INTRODUCTION
LECTURE

Dr.-Ing. Dmitrii Dobriborsci

SENSE ESTIMATE DECIDE
FUSE CONTROL
Critic

...
Actor

person
person car PREDICT Safety
module
PREVENT
Content

Block 1: Course description

Block 2: Introduction & motivation

Block 3: Discussion
Block 1: Course description
Organization

• Classes on Mondays: 14:00 – 15:30, 15:45 – 17:15

• 4 SWS
• 60 hours of attendance
• 60 hours of self-study
• Exam: 150 minutes
Goals of this course

• provide an overview of fundamentals algorithms for

mobile robots
• dealing with noisy data
• hands-on experience
Syllabus

Content

Prerequisites • Intro & Motivation

• linear algebra • Mobile robots and

kinematics
• basic physics
• Perception
• basic programming skills
(Python and/or C++) • Localization

• control systems • SLAM

engineering • Motion planning
• EXTRAS: AI for robotics
Block 2: Introduction & motivation
Introduction & motivation

Problem 1
Tasks addressed that need to be solved by robots?

Solution to Problem 1
• Navigation

• Perception

• Learning

• Cooperation

• Acting

• Interaction

• Manipulation & Grasping

• Planning

• etc.
Introduction & Motivation

Robotics yesterday
• Highly repeatable tasks
• Robots bolted to a ground
• Limited to no perception
• Very little AI
Figure 1: Industrial robots

Image source: https://upload.wikimedia.org/

Introduction & Motivation

Current trends
• Entertainment, toys
• Service
• Medical, surgery
• Industrial automation
• Exploration at hazardous
area
• Self-driving cars
• etc.

Image source: Ericsson, Wikimedia, Alcimed, Mobile Insustrial Robot

The key questions

• The three main questions

in Mobile robotics
• Where I am?
• Where am I going? Solution to Problem 2
• How do I get there? • have a model of the environment (a
priori or autonomously built)
Problem 2
To answer these questions the robot • perceive and analyze the

must ... environment

• find its position within the

environment

• plan and execute motions

The see-think-act cycle

knowledge, Mission
database commands
/

Position"
"Global Cognition.
Localization
Map building map Path Planning

Environment model
Local map Path

7
Information Path execution
extraction and
interpretation See-think-act
M
É
Ram data Actuator cards
÷:
ed
Sensine Acting

Real world
Environment
Motion control | kinematics

Wheel types and its constraints

• Rolling constraints
• No-sliding constraints

 
ẋ
 
ẏ  = f (φ̇1 , . . . , φ̇n , θ, geometry)
Xp
θ̇ y

x
 
φ̇1
 . 
 ..  = f (ẋ, ẏ, θ) y

 
φ̇n "
Autonomous robotics | the see-think-act cycle

knowledge, Mission
database commands
/

Position"
"Global Cognition.
Localization
Map building map Path Planning

Environment model
Local map Path

7
Information Path execution
extraction and
interpretation See-think-act
M
É
Ram data Actuator cards
÷:
ed
Sensine Acting

Real world
Environment
Perception | sensing

lens
object focal

plane

Image sources: Velodyne, https://pjreddie.com/darknet/yolo/, David Paul Morris/Bloomberg/Getty Images

Perception | data extracting

Keypoint features and

matching:
• features that are
reasonably invariant to
rotation, scaling,
viewpoint, illumination
• BRISK, ORB, SIFT, SURF,
FAST, BRIEF ...
Sources: E. Rublee, et. Al., ”ORB: An efficient alternative to SIFT or SURF,” 2011 International Conference on Computer Vision, 2011, pp. 2564-2571, doi:
10.1109/ICCV.2011.6126544.
Perception | understanding

position
Localization (map) Coonition

Env. model path

local map ①

Real Mo ion
Perception
world control

Places/situation < Functional/Contextual

pose, learn, spatial, semantic...

Servicing

Objects
< Models: pose, learn

interaction
É

Features
É ← Models: pose, learn
Navigation
É
Raw data

Sources: Roland Siegwart, Illah R. Nourbakhsh, and Davide Scaramuzza. 2011. Introduction to Autonomous Mobile Robots (2nd. ed.). The MIT Press.
Mapping

• Functional
scene
recognition
• Robots
position in
the envi-
ronment
• Detecting
recognized
static and
dynamic
obstacles
https://emanual.robotis.com/docs/en/platform/turtlebot3/slam/
Autonomous robotics | the see-think-act cycle

knowledge, Mission
database commands
/

Position"
"Global Cognition.
Localization
Map building map Path Planning

Environment model
Local map Path

7
Information Path execution
extraction and
interpretation See-think-act
M
É
Ram data Actuator cards
÷:
ed
Sensine Acting

Real world
Environment
Localization

• SEE: the robot queries its

sensors and finds itself
next to a pillar
• ACT: robot moves one
meter forward: motion
estimated by wheel
encoders, accumulation of
uncertainty
• SEE: the robot queries its
sensors again and finds
itself next to a pillar
• UPDATE
Sources: ETH Zurich
Cognition

Where am I going? How do I get there?

Sources: ETH Zurich

Cognition

• Global path planning: e.g.

graph-search
• Local path planning: local
collision avoidance

Sources: ETH Zurich

Block 3: Discussion
Future potential

Intelligence, interconnection, collaboration, adaptive, learning

Sources: ETH Zurich

Future potential

Our real “physical” world is multimodal, very diverse and

complex.
We need robots that ...
... can dealing with uncertain
and partially available information
... see, feel and understand their environment
... have torque and force control
for tactile interaction (“soft robots”)
... offer intuitive human-machine interfaces
... learn and adapt every day
!!!all this required AI/ML but even more, novel
sensing, actuation and robot concepts!!
Sources: ETH Zurich