SlideShare a Scribd company logo
1 of 43
Machine Vision and Perception Group @ TUM




                                            Perception in Robotic Applications



                                                       Darius Burschka
                                               Machine Vision and Perception Group
                                                   Department of Informatics
                                                  Technische Universität München
MVP




                                                                      Copyright © 2011 Darius Burschka
Research of the MVP Group                        http://mvp.visual-navigation.com
                                                                               Visual navigation
                                              Perception for manipulation
Machine Vision and Perception Group @ TUM




                                                                                                          Biologically motivated
                                                                                                               perception




                                             Rigid and Deformable                 The Machine
                                                  Registration            Vision and Perception Group
                                                                        @TUM works on the aspects of
                                                                       visual perception and control in
                                                                           medical, mobile, and HCI
                                                                                  applications

                                                                               Photogrammetric            Visual Action Analysis
                                             Exploration of physical        monocular reconstruction
                                               object properties
MVP




                                                                                                   Copyright © 2011 Darius Burschka
Brook's Layered Control Architecture
Machine Vision and Perception Group @ TUM




                                            awareness      complexity/
                                                             speed         Brooks, 1986 introduced the
                                                                          paradigm of layered control to
                                                                          the robotics community,
                                                                          implementing many of the
                                                                          features of Minsky's agents.
                                                                          Rather than decomposing the
                                                                          autonomous control problem
                                                                          into separate layers in a
                                                                          sense/model/plan/act scheme,
                                                                          Brooks advocates an
                                                                          architecture where each layer
                                                                          tightly couples sensing to
                                                                          actuators.
MVP




                                                                         Copyright © 2011 Darius Burschka
“Instincts” Modeling as low-level Action Planner
Machine Vision and Perception Group @ TUM




                                                   Sensors
                                            Camera,IMU,VGPS,Laser   map-based action planning            Actuators




                                                                                                     M
                                                                                                     M
MVP




                                                                          Reactive behavior
                                                                    e.g., Obstacle avoidance,…
                                                                                          Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM




                                             Navigation Alternatives




                                            Map-based Navigation                 Vision-Based Control
                                              data is stored and fused in 3D       direct coupling of perception to
                                              representations which are useful     action
MVP




                                              for mission planning
                                                                                        Οi = ℑ−1 ⋅ Ιi


                                                                                       Copyright © 2011 Darius Burschka
Visual Localization for hybrid Environmental Modeling
Machine Vision and Perception Group @ TUM




                                            Real-Time Localization

                                            The high accuracy allows direct
                                            stiching of images along the
                                            trajectory without bundle
                                            adjustment

                                            Resulting hybrid (appearance
                                            and geometry) model allows
                                            path planning and prediction of
                                            sensor views for e.g. attention
MVP




                                            research


                                                                              Copyright © 2011 Darius Burschka
Example of a hybrid model reconstruction
Machine Vision and Perception Group @ TUM
MVP




                                                                        Copyright © 2011 Darius Burschka
Result of a visual localization
Machine Vision and Perception Group @ TUM




                                             6DoF pose of the robot is calculated fom single
                                             camera view
MVP




                                                                             Copyright © 2011 Darius Burschka
Reconstruction of a hybrid (appearance/geometry) model
Machine Vision and Perception Group @ TUM
MVP




                                                                                    Copyright © 2011 Darius Burschka
Visual Localization for
                                            Visual Environment Modeling(collaboration with Meier/Steinbach)]
Machine Vision and Perception Group @ TUM




                                                         localization result




                                                                                                                           surprise
                                                                                                                           trigger
MVP




                                                                                                w


                                                                                                        10
                                                                                                    Copyright © 2011 Darius Burschka
Per-pixel Depth Maps
Machine Vision and Perception Group @ TUM




                                            Pixel get an additional attribute -> the depth
MVP




                                                                                 Copyright © 2011 Darius Burschka
Visual Homing
Machine Vision and Perception Group @ TUM




                                            The system is capable of
                                            registration to previously
                                            observed trajectories
MVP




                                                                         Copyright © 2011 Darius Burschka
Motivation
Machine Vision and Perception Group @ TUM




                                            global registration without
                                              external devices
MVP




                                                                                    13
                                                                          Copyright © 2011 Darius Burschka
Video: Motion Model + IMU Support
Machine Vision and Perception Group @ TUM




                                                                                          Video Link
MVP




                                                                              14
                                                                    Copyright © 2011 Darius Burschka
Stereo Initialization
Machine Vision and Perception Group @ TUM




                                            KLT based stereo initialization
MVP




                                                                                        15
                                                                              Copyright © 2011 Darius Burschka
Feature Set Management
Machine Vision and Perception Group @ TUM




                                                                          Video Link
MVP




                                                                          16
                                                                Copyright © 2011 Darius Burschka
Embedded Vision at an Example of a Flying Application
Machine Vision and Perception Group @ TUM
MVP




                                                                                   Copyright © 2011 Darius Burschka
Work on Optimal Sensor Models
Machine Vision and Perception Group @ TUM




                                                                                       Video Link
MVP




                                                                     Copyright © 2011 Darius Burschka
New Vision System– with Canberra National University,
                                            Australia and University Bielefeld
Machine Vision and Perception Group @ TUM
MVP




                                                                                 Copyright © 2011 Darius Burschka
Biologically Motivated Navigation
Machine Vision and Perception Group @ TUM




                                                      Obstacle avoidance




                                                                                             Decoupling of Rotation
                                                                                             and Translation
MVP




                                                         Perception-based loop closure




                                                                                   Copyright © 2011 Darius Burschka
Direct Localization
Machine Vision and Perception Group @ TUM




                                                Direct calculation of rotation
MVP




                                                                         Copyright © 2011 Darius Burschka
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
„Simple“ Image Acquisition
Machine Vision and Perception Group @ TUM




                                              60 images taken with a standard low cost digital camera
MVP




                                                                                           Copyright © 2011 Darius Burschka
Estimation of the 6 Degrees of
                                            Freedom
Machine Vision and Perception Group @ TUM




                                            Estimation of 3 rotational angles   Estimation of a translation vector
MVP




                                                                                      Copyright © 2011 Darius Burschka
Action parsing
Machine Vision and Perception Group @ TUM
MVP




                                                             Copyright © 2011 Darius Burschka
Perception for manipulation
                                            Example of the Manipulation System in Action
Machine Vision and Perception Group @ TUM




                                                                                                 Video Link
MVP




                                                                           Copyright © 2011 Darius Burschka
Knowledge Representation
Machine Vision and Perception Group @ TUM




                                                                               Human demonstration

                                            • Atlas:                                                                 Atlas
                                                                                         parametric     grasp      actions and
                                               – Long-term memory                shape   description    points      handling

                                               – Experience of the system
                                            • Working memory:
                                               – Short-term memory
                                                                                                   Mapping of
                                               – Experience grounded in a                          Knowledge

                                                 given environment            Foreground                                Experience
                                                                              segmentation                              abstraction
                                            • Temporal handling information       background    foreground       Episodic buffer
                                                                                     model          model          of actions
MVP




                                                                                                                 Working Memory




                                                                                    Copyright © 2011 Darius Burschka
Knowledge Representation
Machine Vision and Perception Group @ TUM




                                                           Mapping of
                                                           Knowledge
MVP




                                                                Copyright © 2011 Darius Burschka
Grounding of Actions: Ontology→Working Memory
Machine Vision and Perception Group @ TUM
MVP




                                                                       Copyright © 2011 Darius Burschka
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
Action parsing
Machine Vision and Perception Group @ TUM
MVP




                                                             Copyright © 2011 Darius Burschka
Action parsing
Machine Vision and Perception Group @ TUM
MVP




                                                             Copyright © 2011 Darius Burschka
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
Copyright © 2011 Darius Burschka
Machine Vision and Perception Group @ TUM   MVP
Basic Experiments:
Machine Vision and Perception Group @ TUM




                                            Functionality Maps (Tracking Data)




                                             Milk carton
                                                                              Cup - Handle
MVP




                                                    Red = goal, green = push, magenta = arbitrary

                                                                                         Copyright © 2011 Darius Burschka
Estimation of Physical Properties
                                            in Predict-Act-Perceive loop
Machine Vision and Perception Group @ TUM




                                                   Estimation of the Center of mass
                                                   Estimation of Mass and Friction Force
                                                   Estimation of Mass Distribution

                                            predict                 act                    perceive

                                               F




                                                       r



                                                           τ
MVP




                                            simulator


                                                                                             Copyright © 2011 Darius Burschka
How to reconstruct 3D under poor texture
                                            condition?
Machine Vision and Perception Group @ TUM




                                              Problem: texture information is more sparse
MVP




                                                                                   39
                                                                            Copyright © 2011 Darius Burschka
What can we do if the texture information is
                                            almost non-existent?
Machine Vision and Perception Group @ TUM




                                            → photogrammetric approach
MVP




                                                                                40
                                                                         Copyright © 2011 Darius Burschka
Reconstruction Example
Machine Vision and Perception Group @ TUM




                                            Works well under static lighting conditions and
                                             roughly Lambertian surfaces
MVP




                                               Ruepp and Burschka. Fast recovery of weakly textured surfaces from monocular image
                                                                                                             sequences. ACCV 2010

                                                                                                Copyright © 2011 Darius Burschka
Research of the MVP Group                        http://mvp.visual-navigation.com
                                                                               Visual navigation
                                              Perception for manipulation
Machine Vision and Perception Group @ TUM




                                                                                                          Biologically motivated
                                                                                                               perception




                                             Rigid and Deformable                 The Machine
                                                  Registration            Vision and Perception Group
                                                                        @TUM works on the aspects of
                                                                       visual perception and control in
                                                                           medical, mobile, and HCI
                                                                                  applications

                                                                               Photogrammetric            Visual Action Analysis
                                             Exploration of physical        monocular reconstruction
                                               object properties
MVP




                                                                                                   Copyright © 2011 Darius Burschka
Conclusions
Machine Vision and Perception Group @ TUM




                                            Perception needs to be coupled to technical
                                            systems at the lowest possible level to avoid
                                            unnecessary error accumulation in the
                                            abstraction process

                                            Perception is not only based on texture and
                                            stereo

                                            Sensor-fusion is important for robust perception
MVP




                                                                            Copyright © 2011 Darius Burschka

More Related Content

Similar to Bosch Expert Days

Sustainable in asset management
Sustainable in asset managementSustainable in asset management
Sustainable in asset managementAdiirfan Che-ani
 
#learning3 #mobilelearning
#learning3 #mobilelearning#learning3 #mobilelearning
#learning3 #mobilelearningAaron Ripin
 
3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop
3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop
3D context-aware mobile maps for tourism - ENTER2011 PhD WorkshopZornitza Yovcheva
 
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...Kalle
 
ex of my work for client: Corporate Back Office suite On Cloud
ex of my work for client: Corporate Back Office suite On Cloudex of my work for client: Corporate Back Office suite On Cloud
ex of my work for client: Corporate Back Office suite On CloudPravin Gandhi
 
Auto DCR building plan approval-nov2012
Auto DCR building plan approval-nov2012Auto DCR building plan approval-nov2012
Auto DCR building plan approval-nov2012SoftTech Engineers
 
Mobile Cloud Computing
Mobile Cloud ComputingMobile Cloud Computing
Mobile Cloud ComputingDejan Kovachev
 
101206 intelligent buildings
101206 intelligent buildings101206 intelligent buildings
101206 intelligent buildingsMartijn Moerbeek
 
Brokerage 2007 presentation image
Brokerage 2007 presentation imageBrokerage 2007 presentation image
Brokerage 2007 presentation imageimec.archive
 
The Concept of Sensing as a Service Using Mobile Crowdsensing
The Concept of Sensing as a Service Using Mobile CrowdsensingThe Concept of Sensing as a Service Using Mobile Crowdsensing
The Concept of Sensing as a Service Using Mobile CrowdsensingDr. Mazlan Abbas
 

Similar to Bosch Expert Days (16)

Mvp
MvpMvp
Mvp
 
Sustainable in asset management
Sustainable in asset managementSustainable in asset management
Sustainable in asset management
 
#learning3 #mobilelearning
#learning3 #mobilelearning#learning3 #mobilelearning
#learning3 #mobilelearning
 
3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop
3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop
3D context-aware mobile maps for tourism - ENTER2011 PhD Workshop
 
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...
Porta Ce Cursor A Contextual Eye Cursor For General Pointing In Windows Envir...
 
Apmac poster 3 collums
Apmac poster 3 collumsApmac poster 3 collums
Apmac poster 3 collums
 
ex of my work for client: Corporate Back Office suite On Cloud
ex of my work for client: Corporate Back Office suite On Cloudex of my work for client: Corporate Back Office suite On Cloud
ex of my work for client: Corporate Back Office suite On Cloud
 
Auto DCR building plan approval-nov2012
Auto DCR building plan approval-nov2012Auto DCR building plan approval-nov2012
Auto DCR building plan approval-nov2012
 
Mobile Cloud Computing
Mobile Cloud ComputingMobile Cloud Computing
Mobile Cloud Computing
 
visionCHANGE
visionCHANGEvisionCHANGE
visionCHANGE
 
101206 intelligent buildings
101206 intelligent buildings101206 intelligent buildings
101206 intelligent buildings
 
Mobile device management
Mobile device management Mobile device management
Mobile device management
 
Webinar: Mobile Device Management
Webinar: Mobile Device ManagementWebinar: Mobile Device Management
Webinar: Mobile Device Management
 
Stream 1 - Cloud Computing
Stream 1 - Cloud ComputingStream 1 - Cloud Computing
Stream 1 - Cloud Computing
 
Brokerage 2007 presentation image
Brokerage 2007 presentation imageBrokerage 2007 presentation image
Brokerage 2007 presentation image
 
The Concept of Sensing as a Service Using Mobile Crowdsensing
The Concept of Sensing as a Service Using Mobile CrowdsensingThe Concept of Sensing as a Service Using Mobile Crowdsensing
The Concept of Sensing as a Service Using Mobile Crowdsensing
 

Bosch Expert Days

  • 1. Machine Vision and Perception Group @ TUM Perception in Robotic Applications Darius Burschka Machine Vision and Perception Group Department of Informatics Technische Universität München MVP Copyright © 2011 Darius Burschka
  • 2. Research of the MVP Group http://mvp.visual-navigation.com Visual navigation Perception for manipulation Machine Vision and Perception Group @ TUM Biologically motivated perception Rigid and Deformable The Machine Registration Vision and Perception Group @TUM works on the aspects of visual perception and control in medical, mobile, and HCI applications Photogrammetric Visual Action Analysis Exploration of physical monocular reconstruction object properties MVP Copyright © 2011 Darius Burschka
  • 3. Brook's Layered Control Architecture Machine Vision and Perception Group @ TUM awareness complexity/ speed Brooks, 1986 introduced the paradigm of layered control to the robotics community, implementing many of the features of Minsky's agents. Rather than decomposing the autonomous control problem into separate layers in a sense/model/plan/act scheme, Brooks advocates an architecture where each layer tightly couples sensing to actuators. MVP Copyright © 2011 Darius Burschka
  • 4. “Instincts” Modeling as low-level Action Planner Machine Vision and Perception Group @ TUM Sensors Camera,IMU,VGPS,Laser map-based action planning Actuators M M MVP Reactive behavior e.g., Obstacle avoidance,… Copyright © 2011 Darius Burschka
  • 5. Machine Vision and Perception Group @ TUM Navigation Alternatives Map-based Navigation Vision-Based Control data is stored and fused in 3D direct coupling of perception to representations which are useful action MVP for mission planning Οi = ℑ−1 ⋅ Ιi Copyright © 2011 Darius Burschka
  • 6. Visual Localization for hybrid Environmental Modeling Machine Vision and Perception Group @ TUM Real-Time Localization The high accuracy allows direct stiching of images along the trajectory without bundle adjustment Resulting hybrid (appearance and geometry) model allows path planning and prediction of sensor views for e.g. attention MVP research Copyright © 2011 Darius Burschka
  • 7. Example of a hybrid model reconstruction Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 8. Result of a visual localization Machine Vision and Perception Group @ TUM 6DoF pose of the robot is calculated fom single camera view MVP Copyright © 2011 Darius Burschka
  • 9. Reconstruction of a hybrid (appearance/geometry) model Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 10. Visual Localization for Visual Environment Modeling(collaboration with Meier/Steinbach)] Machine Vision and Perception Group @ TUM localization result surprise trigger MVP w 10 Copyright © 2011 Darius Burschka
  • 11. Per-pixel Depth Maps Machine Vision and Perception Group @ TUM Pixel get an additional attribute -> the depth MVP Copyright © 2011 Darius Burschka
  • 12. Visual Homing Machine Vision and Perception Group @ TUM The system is capable of registration to previously observed trajectories MVP Copyright © 2011 Darius Burschka
  • 13. Motivation Machine Vision and Perception Group @ TUM global registration without external devices MVP 13 Copyright © 2011 Darius Burschka
  • 14. Video: Motion Model + IMU Support Machine Vision and Perception Group @ TUM Video Link MVP 14 Copyright © 2011 Darius Burschka
  • 15. Stereo Initialization Machine Vision and Perception Group @ TUM KLT based stereo initialization MVP 15 Copyright © 2011 Darius Burschka
  • 16. Feature Set Management Machine Vision and Perception Group @ TUM Video Link MVP 16 Copyright © 2011 Darius Burschka
  • 17. Embedded Vision at an Example of a Flying Application Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 18. Work on Optimal Sensor Models Machine Vision and Perception Group @ TUM Video Link MVP Copyright © 2011 Darius Burschka
  • 19. New Vision System– with Canberra National University, Australia and University Bielefeld Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 20. Biologically Motivated Navigation Machine Vision and Perception Group @ TUM Obstacle avoidance Decoupling of Rotation and Translation MVP Perception-based loop closure Copyright © 2011 Darius Burschka
  • 21. Direct Localization Machine Vision and Perception Group @ TUM Direct calculation of rotation MVP Copyright © 2011 Darius Burschka
  • 22. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 23. „Simple“ Image Acquisition Machine Vision and Perception Group @ TUM 60 images taken with a standard low cost digital camera MVP Copyright © 2011 Darius Burschka
  • 24. Estimation of the 6 Degrees of Freedom Machine Vision and Perception Group @ TUM Estimation of 3 rotational angles Estimation of a translation vector MVP Copyright © 2011 Darius Burschka
  • 25. Action parsing Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 26. Perception for manipulation Example of the Manipulation System in Action Machine Vision and Perception Group @ TUM Video Link MVP Copyright © 2011 Darius Burschka
  • 27. Knowledge Representation Machine Vision and Perception Group @ TUM Human demonstration • Atlas: Atlas parametric grasp actions and – Long-term memory shape description points handling – Experience of the system • Working memory: – Short-term memory Mapping of – Experience grounded in a Knowledge given environment Foreground Experience segmentation abstraction • Temporal handling information background foreground Episodic buffer model model of actions MVP Working Memory Copyright © 2011 Darius Burschka
  • 28. Knowledge Representation Machine Vision and Perception Group @ TUM Mapping of Knowledge MVP Copyright © 2011 Darius Burschka
  • 29. Grounding of Actions: Ontology→Working Memory Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 30. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 31. Action parsing Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 32. Action parsing Machine Vision and Perception Group @ TUM MVP Copyright © 2011 Darius Burschka
  • 33. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 34. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 35. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 36. Copyright © 2011 Darius Burschka Machine Vision and Perception Group @ TUM MVP
  • 37. Basic Experiments: Machine Vision and Perception Group @ TUM Functionality Maps (Tracking Data) Milk carton Cup - Handle MVP Red = goal, green = push, magenta = arbitrary Copyright © 2011 Darius Burschka
  • 38. Estimation of Physical Properties in Predict-Act-Perceive loop Machine Vision and Perception Group @ TUM Estimation of the Center of mass Estimation of Mass and Friction Force Estimation of Mass Distribution predict act perceive F r τ MVP simulator Copyright © 2011 Darius Burschka
  • 39. How to reconstruct 3D under poor texture condition? Machine Vision and Perception Group @ TUM Problem: texture information is more sparse MVP 39 Copyright © 2011 Darius Burschka
  • 40. What can we do if the texture information is almost non-existent? Machine Vision and Perception Group @ TUM → photogrammetric approach MVP 40 Copyright © 2011 Darius Burschka
  • 41. Reconstruction Example Machine Vision and Perception Group @ TUM Works well under static lighting conditions and roughly Lambertian surfaces MVP Ruepp and Burschka. Fast recovery of weakly textured surfaces from monocular image sequences. ACCV 2010 Copyright © 2011 Darius Burschka
  • 42. Research of the MVP Group http://mvp.visual-navigation.com Visual navigation Perception for manipulation Machine Vision and Perception Group @ TUM Biologically motivated perception Rigid and Deformable The Machine Registration Vision and Perception Group @TUM works on the aspects of visual perception and control in medical, mobile, and HCI applications Photogrammetric Visual Action Analysis Exploration of physical monocular reconstruction object properties MVP Copyright © 2011 Darius Burschka
  • 43. Conclusions Machine Vision and Perception Group @ TUM Perception needs to be coupled to technical systems at the lowest possible level to avoid unnecessary error accumulation in the abstraction process Perception is not only based on texture and stereo Sensor-fusion is important for robust perception MVP Copyright © 2011 Darius Burschka