Scribd
Upload
39 views

Autonomous Mobile Robots: Zürich

This document provides an introduction to autonomous mobile robotics. It discusses key concepts like localization, mapping, and path planning. Environment representations can be topological, metric, or a combination. Probabilistic techniques like Kalman filters and Markov localization are important for localization and mapping in uncertain environments. The document presents examples of state-of-the-art mobile robots including PR2, ASIMO, Google's autonomous vehicles, and swarm of micro flying robots. It also discusses projects involving groups from several universities applying these concepts.

Uploaded by

mbede
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
39 views

Autonomous Mobile Robots: Zürich

This document provides an introduction to autonomous mobile robotics. It discusses key concepts like localization, mapping, and path planning. Environment representations can be topological, metric, or a combination. Probabilistic techniques like Kalman filters and Markov localization are important for localization and mapping in uncertain environments. The document presents examples of state-of-the-art mobile robots including PR2, ASIMO, Google's autonomous vehicles, and swarm of micro flying robots. It also discusses projects involving groups from several universities applying these concepts.

Uploaded by

mbede
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 36

Zrich

Roland Siegwart Margarita Chli Martin Rufli Davide Scaramuzza

ETH Master Course: 151-0854-00L

Autonomous Mobile Robots

Introduction

Key Questions and Concepts in Autonomous Mobile Robotics


The three key questions in Mobile Robotics
Where am I ? Where am I going ? How do I get there ?

To answer these questions the robot has to


have a model of the environment (given or autonomously built) perceive and analyze the environment find its position/situation within the environment plan and execute the movement
Zrich

1 - Introduction

Generic Control Scheme for Mobile Robot Systems


knowledge, data base mission commands

Localization Map Building environment model local map Information Extraction raw data

position global map

Cognition Path Planning path Path Execution actuator commands

Sensing

Acting

Real World Environment


1 - Introduction
Zrich

Motion Control

Perception

Control Architectures / Strategies Control Loop


dynamically changing environment no compact model available many sources of uncertainties
Localization position global map

Two Approaches
Classical AI
complete modeling model based horizontal decomposition

Cognition

environment model local map

path Motion Control

New AI, AL
sparse or no modeling behavior based vertical decomposition bottom up
Zrich

Perception

Real World Environment

1 - Introduction

Mixed Approach Depicted into the Generic Control Scheme

Localization

position / global map position / global map local map local map
perception to action

Cognition

obstacle avoidance

environment model local map

position feedback

Perception
1 - Introduction

Real World Environment

Motion Control

path
Zrich

Environment Representation and Modeling

1 - Introduction

Zrich

Environment Representation and Modeling: How we do it


Odometry
not applicable

Modified Environments
expensive, inflexible
39

Feature-based Navigation
121 95

34

still a challenge for artificial systems

25

Elevator door

Corridor crossing

How to find a treasure Landing at night

Entrance Eiffel Tower


1 - Introduction
Zrich

Human Navigation: Topology with rough metric information

~ 400 m

~ 1 km ~ 200 m

~ 50 m ~ 10 m

1 - Introduction

Zrich

Environment Representation: The Map Categories


Topological Maps (Recognizable Locations)

Metric Topological Maps

Fully Metric Maps (continuous or


discrete)
y
200 m

50 km 2 km

100 km

{W}
1 - Introduction

x
Zrich

Courtesy K. Arras

10

From Perception to Understanding

Places / Situations
A specific room, a meeting situation,

Fusing & Compressing Information

Servicing / Reasoning

Objects
Doors, Humans, Coke bottle, car ,

Functional / Contextual Relationships of Objects


imposed learned spatial / temporal/semantic

Interaction

Models / Semantics

Features
Lines, Contours, Colors, Phonemes,

imposed learned

Navigation

Models

Raw Data
Vision, Laser, Sound, Smell,

imposed learned

1 - Introduction

Zrich

11

Understanding Probabilistic Reasoning (e.g. Bayesian)


Reasoning in the presence of uncertainties and incomplete information Combining preliminary information and models with learning from experimental data

Picture Courtesy of Bessiere, INRIA Grenoble, France

1 - Introduction

Zrich

12

Improving belief state by moving SEE and ACT

The robot is placed somewhere in the environment


Location unknown
x

The robot queries its sensors (SEE)


finds itself next to a pillar

The robot moves one meter forward (ACT)


Motion estimated by wheel encoders Accumulation of uncertainty

The robot queries its sensors again (SEE)


finds itself again next to a pillar

Updates its belief by combining this information with its previous belief
1 - Introduction
Zrich

13

Metric Navigation: Probabilistic Position Estimation


Kalman Filter Localization
Continuous, recursive and very compact

SEE
Matching

ACT

1 - Introduction

Zrich

14

State Prediction through Odometry


Incrementally (dead reckoning)
Odometry (wheel encoder) and/or initial sensors (gyro) Drift -> increasing uncertainty

ACT

1 - Introduction

Zrich

15

Measurement Prediction and Observation


Measurement Prediction
Using state prediction and e.g. metric map

Observation
e.g. feature (line segments) extracted from laser scan

1 - Introduction

Zrich

16

Matching and Estimation


Matching:
Find correspondence of features
Observation

Prediction

Estimation (new position):


Weighted mean (prediction & observation) e.g. Kalman filter, Markov

Observation

Prediction (Odometry)

1 - Introduction

Zrich

17

Different Approaches to Belief Representation


a. Continuous map with single hypothesis
(e.g. Kalman Filter)

b. Continuous map with multiple hypothesis


(e.g. Multiple Kalman Filters)

c. Discretized map with probability distribution


(e.g. Markov Localization)

d. Discretized topological map with probability distribution


1 - Introduction
Zrich

18

Discretizes Map Grid-Based Metric Approach


Grid Map of the Smithsonians National Museum of American History in Washington DC. Markov Localization Grid: ~ 400 x 320 = 128000 points

1 - Introduction

Courtesy S. Thrun, W. Burgard

Zrich

19

Grid-Based SLAM
(Simultaneous Localization and Mapping) Partial Filter to reduce computational complexity

Courtesy of Sebastian Thrun


1 - Introduction
Zrich

20

Probabilistic 3D SLAM

raw data
raw 3D scan of the same scene

find a plane for every cell using RANSAC

fuse similar neighboring planes together

segmented planar segments


1 - Introduction
final segmentation
Zrich

one plane per grid cell

photo of the scene

decompose space into grid cells fill cells with data

21

Probabilistic 3D SLAM

close-up of reconstructed bookshelves close-up of a reconstructed hallway

Incremental map-building using a probabilistic estimation process. magenta: odometry only; blue: SLAM
1 - Introduction
Zrich

22

Data Compression: 99%


J. Weingarten
along its way (140 m), the robot takes 90 3D scans; the total number of planar segments is 244 (44696 data points / 299 polygons). This corresponds to a compression ratio of more than 99% w.r.t. raw data (5212800 points).

the robot lacks sensors to estimate 3D trajectories ICP or laser-corrected odometry allows to simulate a 6D odometry. This makes reconstruction of nonflat environments possible

1 - Introduction

Zrich

23

Maps for Planning: Exploration and Graph Construction


1. Exploration
explore on stack already examined

2. Graph Construction

provides correct topology must recognize already visited location backtracking for unexplored openings
1 - Introduction

Where to put the nodes?


Topological: at distinctive locations Metric-based: where features disappear or get visible
Zrich

24

Some examples of todays state-of-the-art mobile robots

1 - Introduction

Zrich

25

PR2: Personal Robot 2


Robot for research and experimentation Development platform:
Cameras, Laser scanners, Accelerometer, Tactile sensors 16 CPU cores Sophisticated joints design for safety Variety of networking tools for communicating data
Inside the PR2

ROS: Robot Operating System free, open source, software development platform integrating libraries and tools Cost: $400 000

Courtesy of Willow Garage


1 - Introduction
Zrich

26

PR2: applications

Fold towels

Courtesy of Clean-up with cart


1 - Introduction
Zrich

27

Humanoid Robot: ASIMO


Hondas ASIMO - Advanced Step in Innovative MObility Designed to help people in their everyday lives One of the most advanced humanoid robots
Compact, lightweight Sophisticated walk technology Human-friendly design

Video: Honda 2012

1 - Introduction

Zrich

28

Google: Autonomous Driving in traffic


October 2010 Self-driving car in real traffic Toyota Prius + a variety of sensors:
Lidar (Laser) Video camera Radars GPS receiver Wheel encoders etc.

Autonomous Driving:
sense the surroundings mimic the decisions of a human driver

1 - Introduction

Zrich

29

Google: Autonomous Driving in traffic


Plan route like a GPS navigator but use extra data to decide on driving actions Boost safety & efficiency 7 cars,140000 miles with minimal human intervention Autonomous cars are still years from mass production

Video: ABC News


1 - Introduction
Zrich

30

EUROPA
European Robotic Pedestrian Assistant
The Team: ETH Zurich, University of Freiburg, Univ. of Oxford, KU Leuven, RWTH Aachen, BlueBotics

Zrich

31

NIFTi Urban Search and Rescuing


Project goals Robotic help for Urban Search and Rescue UGV and UAV combined for scene exploration Yearly evaluation of system by firemen Environment modeling Online 3D mapping from laser sensor Based on enhanced ICP released open-source Topological segmentation for human-robot interaction
The Team: ETH Zurich, DFKI Saarbrucken, TNO Soesterberg, Fraunhofer St. Augustin, BlueBotics, Czech Technical University, La Sapienza University of Rome, Fire Department of Dortmund, Ministry of the Interior Italy
Zrich

Omnicam Rotating Laser

32

sFly: Swarm of micro Flying Robots


Recently finished EU project Coordinated flight in small swarms over previously unknown areas Autonomous micro helicopters for: Access to dangerous environments:
inspection, exploration, search & rescue, monitoring & surveillance

Vision-only fully autonomous navigation


GPS-denied environments
www.sfly.ethz.ch/

The Team:
ETH Zurich, (Autonomous Systems Lab and Computer Vision and Geometry Group); Ascending Technologies; Technical University of Crete; INRIA Grenoble; CSEM

1 - Introduction

Zrich

33

Feature Traking BRISK

BRISK:
Binary Robust Invariant Scalable Keypoints Performance comparable to SURF/SIFT speed up to 10 times faster than SURF (CPU)

Zrich

34

sFly Swarm of micro Flying robots with feature based visual navigation

Zrich

35

Content of the Course


1. 2. 3. 4. 5. 6. Introduction Locomotion Mobile Robot Kinematics Perception Mobile Robot Localization and Mapping Planning and Navigation

Specific aspect and potential applications are presented along the course

1 - Introduction

Zrich

36

Slides and additional material


Slides and Exercises
http://www.asl.ethz.ch/education/master/mobile_robotics Currently you can find there the last years slides and exercises. This years material will gradually become available throughout the term

Relevant reading material:


Introduction to Autonomous Mobile Robots
Roland Siegwart, Illah Nourbakhsh, Davide Scaramuzza Intelligent Robotics and Autonomous Agents series The MIT Press, Massachusetts Institute of Technology Cambridge, Massachusetts 02142 ISBN 0-262-19502-X

http://www.mobilerobots.org

1 - Introduction

Zrich

You might also like