Sebastian Thrun

1
Probabilistic Algorithms forMobile Robot Mapping

• Sebastian Thrun
• Carnegie Mellon Stanford
• Wolfram Burgard
• University of Freiburg
• and Dieter Fox
• University of Washington

2

• Based on the paper
• A Real-Time Algorithm for Mobile Robot Mapping
• With Applications to Multi-Robot and 3D Mapping
• Best paper award at 2000 IEEE International
  Conference on Robotics
• and Automation (1,100 submissions)
• Sponsored by DARPA (TMR-J.Blitch, MARS-D.Gage,
  MICA-S.Heise)
• and NSF (ITR(2), CAREER-E.Glinert,
  IIS-V.Lumelsky)
• Other contributors Yufeng Liu, Rosemary Emery,
  Deepayan Charkrabarti, Frank Dellaert, Michael
• Montemerlo, Reid Simmons, Hugh Durrant-Whyte,
  Somajyoti Majnuder, Nick Roy, Joelle Pineau,

3
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
4
Museum Tour-Guide Robots
With Greg Armstrong, Michael Beetz, Maren
Benewitz, Wolfram Burgard, Armin Cremers, Frank
Dellaert, Dieter Fox, Dirk Haenel, Chuck
Rosenberg, Nicholas Roy, Jamie Schulte, Dirk
Schulz
5
The Nursebot Initiative
With Greg Armstrong, Greg Baltus, Jacqueline
Dunbar-Jacob, Jennifer Goetz, Sara Kiesler,
Judith Matthews, Colleen McCarthy, Michael
Montemerlo, Joelle Pineau, Martha Pollack,
Nicholas Roy, Jamie Schulte
6
(No Transcript)
7
Mapping The Problem

• Concurrent Mapping and Localization (CML)
• Simultaneous Localization and Mapping (SLAM)

8
Mapping The Problem

• Continuous variables
• High-dimensional (eg, 1,000,000 dimensions)
• Multiple sources of noise
• Simulation not acceptable

9
Milestone Approaches
Mataric 1990
Elfes/Moravec 1986
Kuipers et al 1991
Lu/Milios/Gutmann 1997
10
3D Mapping
Moravec et al, 2000
Konolige et al, 2001
Teller et al, 2000
11
Take-Home Message

• Mapping is the
• holy grail in
• mobile robotics.

12
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
13
Bayes Filters
Special cases HMMs DBNs POMDPs Kalman
filters Condensation ...
x state t time z measurement u
control ? constant
14
Bayes Filters in Localization
Simmons/Koenig 95 Kaelbling et al
96 Burgard, Fox, et al 96
15
Bayes Filters for Mapping
s robot pose m map t time ? constant z
measurement u control
16
Kalman Filters (SLAM)
Smith, Self, Cheeseman, 1990
17
Underwater Mapping with SLAMCourtesy of Hugh
Durrant-Whyte, Univ of Sydney
18
Large-Scale SLAM MappingCourtesy of John
Leonard, MIT
19
SLAM Limitations

• Linear
• Scaling O(N4) in number of features in map
• Cant solve data association problem

20
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
21
Unknown Data Association EM
Dempster et al, 77 Thrun et al, 1998
Shatkay/Kaelbling 1997
22
CMUs Wean Hall (80 x 25 meters)
23
EM Mapping, Example (width 45 m)
24
EM Mapping Limitations

• Local Minima
• Not Real-Time

25
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
26
The Goal
EM data association Not real-time
Kalman filters real-time No data association
?
27
Real-Time Approximation (ICRA paper)
28
Incremental ML Not A Good Idea
mismatch
path
robot
29
Real-Time Approximation
Our ICRA Paper ?
30
Real-Time Approximation
Yellow flashes artificially distorted map (30
deg, 50 cm)
31
Importance of Posterior Pose Estimate
Without pose posterior
With pose posterior
32
Online Mapping with PosteriorCourtesy of Kurt
Konolige, SRI, DARPA-TMR
Gutmann Konolige, 00
33
Accuracy The Tech Museum, San Jose
CAD map
2D Map, learned
34
Multi-Robot Mapping
Cascaded architecture
map


map
Aligned map
Pre-aligned scans
map
map
map

• Every module maximizes likelihood
• Pre-aligned scans are passed up in hierarchy

35
Multi-Robot Exploration
DARPA TMR Maryland 7/00
DARPA TMR Texas 7/99 (July. Texas. No air
conditioning. Req to dress up. Rattlesnakes)
36
3D Volumetric Mapping
37
3D Structure Mapping
38
3D Texture Mapping
39
Fine-Grained StructureCan We Do Better?
40
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
41
Multi-Planar 3D Mapping

• Idea Exploit fact that buildings posses many
  planar surfaces
• Compact models
• High Accuracy
• Objects instead of pixels

42
3D Multi-Plane Mapping Problem

• Entails five problems
• Generative model with priors Not all of the
  world is planar
• Parameter estimation Location and angle of
  planar surfaces unknown
• Outlier identification Not all measurements
  correspond to planar surfaces (other objects,
  noise)
• Correspondence Different measurements correspond
  to different planar surfaces
• Model selection Number of planar surfaces unknown

43
Expected Log-Likelihood Function
Liu et al, ICML-01
44
EM To The Rescue!
Game Over!
45
Results
With EM (95 of data explained by 7 surfaces)
Without EM
error
With Deepayan Chakrabarti, Rosemary Emery,
Yufeng Liu, Wolfram Burgard, ICML-01
46
The Obvious Next Step
EM for object mapping
EM for concurrent localization
?
47
Underwater Mapping (with University of Sydney)

With Hugh Durrant-Whyte, Somajyoti Majunder,
Marc de Battista, Steve Scheding
48
This Talk
Motivation
SLAM (Kalman filters)
Expectation Maximization
Real Time Hybrid
3D Mapping with EM
Open Problems
49
Take-Home Message

• Mapping is the
• holy grail in
• mobile robotics.

Every state-of-the-art mapping algorithm is
probabilistic.
Sebastian has one cool animation!
50
Open Problems

• 2D Indoor mapping and exploration
• 3D mapping (real-time, multi-robot)
• Object mapping (desks, chairs, doors, )
• Outdoors, underwater, planetary
• Dynamic environments (people, retail stores)
• Full posterior with data association (real-time,
  optimal)

51
Open Problems, cont

• Mapping, localization
• Control/Planning under uncertainty
• Integration of symbolic making
• Human robot interaction
• Literature Pointers
• Robotic Mapping at www.thrun.org
• Probabilistic Robotics AI Magazine 21(4)

52
www.appliedautonomy.com