63
From Vision to Actions towards adaptive & autonomous robots Jürgen ‘Juxi’ Leitner i'stituto dalle molle di studi sull’intelligenza artificiale idsia / supsi / usi

From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

Embed Size (px)

DESCRIPTION

 

Citation preview

Page 1: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

From Vision to Actions towards adaptive & autonomous robots

Jürgen ‘Juxi’ Leitner i'stituto dalle molle di studi sull’intelligenza artificiale

idsia / supsi / usi

Page 2: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

humanoidiCub

Page 3: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

projectIM-CLeVeR

http://robotics.idsia.ch/im-clever/

Page 4: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

integration

http://robotics.idsia.ch/

Page 5: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

overview

a novel way of object segmentation (CGPIP)

learning and teaching spatial perception

integration action-perception sidereactive reaching/graspingimproving perception with (inter-)actions

Page 6: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

!

perceptionvisual

thanks to G. Metta and IIT for this picture

Page 7: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

object manipulationtowards learning

http://robotics.idsia.ch/

Page 8: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

!

objectsdetecting

Page 9: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

!

objectsdetecting

Harding, Leitner, Schmidhuber, 2013Leitner et al., ICDL 2012, IJARS 2012, BICA 2012, CEC 2013

Page 10: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

+ min dilate avg INP INP INP diff

using building blocksOpenCV

+ min dilate avg INP INP INP thresh+ min dilate avg INP INP INP blur+ min dilate avg INP INP INP normalize+ min dilate avg INP INP INP input

Page 11: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

icVision

! icImage* BlueCupFilter::runFilter() { icImage* node43 = InputImages[4]; icImage* node49 = node43->LocalAvg(15); ! icImage* out = node49->threshold(81.532f); return out; }

framework

Page 12: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

embedding domain knowledge

+ min dilate avg INP INP INP OpenCV functions

full images

Page 13: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

cartesian genetic programming

+ min dilate avg INP INP INP

Harding, Leitner, Schmidhuber, 2013Leitner et al., ICDL 2012, IJARS 2012, BICA 2012, CEC 2013

Page 14: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

learningapproach

[Matthews, 1975]

TP… true positive, TN .. true negative!FP… false positive, FN .. false negative

Page 15: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

learningapproach

Page 16: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detection

! icImage GreenTeaBoxDetector::runFilter() { icImage node0 = InputImages[6]; icImage node1 = InputImages[1]; icImage node2 = node0.absdiff(node1); icImage node5 = node2.SmoothBilateral(11); icImage node12 = InputImages[0]; icImage node16 = node12.Sqrt(); icImage node33 = node16.erode(6); icImage node34 = node33.log(); icImage node36 = node34.min(node5); icImage node49 = node36.Normalize(); ! //cleanup ... icImage out = node49.threshold(230.7218f); return out; }

Page 17: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detect

Page 18: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detect

Page 19: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detection

Page 20: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

approachcgp

[Leitner et al, iSAIRAS 2012]

Page 21: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

visualhand detection

[Leitner et al, CEC 2013]

Page 22: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

handsdetecting

[Leitner et al, CEC 2013]

Page 23: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

approachsupervised learning

BUT

Page 24: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

clusteringfeature

saliencymap

Autonomous Approach

[Leitner et al, ICDL/EpiRob 2012]

Page 25: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

presegmentation

Page 26: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

resultscomparing

[Leitner et al, ICDL/EpiRob 2012]

features

CGP-IP

Page 27: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

object manipulationtowards learning

http://robotics.idsia.ch/

Page 28: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

localisation approachescurrent

[P2,1] * F * [P1,1]' = 0

Page 29: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

!

learningspatial perception

Page 30: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

trainingset

9DOF!

iCubpositio

n in the frame!

2/6 per eye

Carte

sian!

Coor

dinate

s

Page 31: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

!

transferringspatial perception

[Leitner et al, IROS 2012]

Page 32: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

setuplearning

Page 33: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

trainingset

9DOF!

iCubpositio

n in the frame!

2/6 per eye

Carte

sian!

Coor

dinate

s

.

.

.~1000

Page 34: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

spatial perception neural network

.!.!.

9DO

F!iC

ubpo

sitio

n in

the

imag

e!2/

6 pe

r eye

Cart

esian!

Coor

dina

tes

!fu

lly c

onne

cted!

!fu

lly c

onne

cted!

.!.!.

Page 35: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

results ANN

-700

-600

-500

-400

-300

-200

-100

0

100

200

300

400

0 200 400 600 800 1000

Y (

mm

)

Sample Index

PredictedExpected

100

150

200

250

300

350

400

450

500

550

600

0 200 400 600 800 1000

X (

mm

)

Sample Index

-1300

-1200

-1100

-1000

-900

-800

-700

-600

-500

-400

-300

-200

0 200 400 600 800 1000

Z (

mm

)

Sample Index

-700

-600

-500

-400

-300

-200

-100

0

100

200

300

400

0 200 400 600 800 1000

Y (

mm

)

Sample Index

PredictedExpected

100

150

200

250

300

350

400

450

500

550

600

0 200 400 600 800 1000

X (

mm

)

Sample Index

-1300

-1200

-1100

-1000

-900

-800

-700

-600

-500

-400

-300

-200

0 200 400 600 800 1000

Z (

mm

)

Sample Index

[Leitner et al, IJCNN 2013]

Page 36: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

results

[Leitner et al, IJARS 2012]

x"="17.81"−"0.0191"v1"+"0.1527"v4"+"0.1378"v7"+"0.0111"v10""""""−"0.0296"v11"−"0.1207"v12"!

y"="1.1242"+"0.1296"v10"+"0.1156"v8"+"0.0170"v0

GPv0 … left camera x coordinate,v1 … left camera y coordinate,

v2 … right camera x coordinate,v3 … right camera y coordinate,

v4-6 … neckv7-9 … eyes v10-12 … torso

simplified caseon the table

Page 37: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

object manipulationtowards learning

http://robotics.idsia.ch/

Page 38: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

MoBeE[Frank, Leitner et al., ICINCO, 2012]

Page 39: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

MoBeEframework [Frank, Leitner et al., ICINCO, 2012]

Page 40: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

[Frank, Leitner et al., ICINCO, 2012]

Page 41: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

motion

robot[Stollenga, Leitner et al, IROS 2013]

Page 42: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

generationmotion

Shak

ey 2

013

Win

ner

Page 43: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

MoBeEv2[Frank, Leitner et al. 2012, 2013]

Page 44: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

hand/armop-space forcing

CSWorld

CSHand

CSR/CSL

[Leitner et al, in prep]

-

Page 45: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

repertoireaction

[Leitner et al, in prep]

Page 46: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

object manipulationtowards learning

http://robotics.idsia.ch/

Page 47: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

operationtele

Page 48: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

hand/armop-space forcing

CSWorld

CSHand

CSR/CSL

[Leitner et al, in prep]

- CSHand’

Page 49: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

teleoperation

Page 50: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

coordinationhand-eye

Page 51: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

model

Page 52: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

http://robotics.idsia.ch/

Page 53: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

manipulation for improved perception

Page 54: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

manipulation actions

Page 55: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

extracting information

Page 56: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

improveddetection

Page 57: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detection

! icImage* BlueCupFilter::runFilter() { icImage* node43 = InputImages[4]; icImage* node49 = node43->LocalAvg(15); ! icImage* out = node49->threshold(81.532f); return out; }

Page 58: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

detection

! icImage* BlueCupFilter::runFilter() { icImage* node0 = InputImages[4].Exp(); icImage* node5 = InputImages[0]; icImage* node16 = node0->Gabor(-8,14,1,13); icImage* node17 = InputImages[4]->LocalAvg(6); icImage* node18 = node16->Laplace(5); icImage* node19 = node5->Sobel(13,9); icImage* node24 = node17->Erode(5); icImage* node28 = node19->Min(node18); icImage* node29 = node28->Min(node24); icImage* node41 = node29->LocalAvg(7); icImage* node49 = node41->LocalMax(7); ! icImage* out = node49->threshold(68.032f); return out; }

resulting

Page 59: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

conclusions

a novel way of object segmentation

learning and teaching spatial perception

integration action-perception sidereactive reaching/graspingimproving perception with (inter-)actions

Page 60: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

S. Harding!M. Frank!

A. Förster!M. Stollenga!

L. Pape!

http://dilbert.com/strips/comic/2013-10-24/

thanks to

Page 61: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

for listeningthank you

[email protected] !http://Juxi.net/projects

Page 62: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

1/2referencesReactive Reaching and Grasping on a Humanoid: Towards Closing the Action-Perception Loop on the iCub. Jürgen Leitner, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. ICINCO 2014. !Improving Robot Vision Models for Object Detection Through Interaction. Jürgen Leitner, Alexander Förster, Jürgen Schmidhuber. WCCI - IJCNN 2014. !Teleoperation of a 7 DOF Humanoid Robot Arm Using Human Arm Accelerations and EMG Signals.. J Leitner, M Luciw, A Förster, J Schmidhuber. iSAIRAS 2014. !Curiosity driven reinforcement learning for motion planning on humanoids. Mikhail Frank, Jürgen Leitner, Alexander Förster, Jürgen Schmidhuber. Frontiers in Neurorobotics, 7:25, 2014. !Task-Relevant Roadmaps: A Framework for Humanoid Motion Planning. M Stollenga, L Pape, M Frank, J Leitner, A Förster, J Schmidhuber. IROS 2013. !ALife in Humanoids: Developing a Framework to Employ Artificial Life Techniques for High-Level Perception and Cognition Tasks on Humanoid Robots. Jürgen Leitner, Simon Harding, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. Alife WS @ ECAL 2013. !Artificial Neural Networks For Spatial Perception: Towards Visual Object Localisation in Humanoid Robots. Jürgen Leitner, Simon Harding, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. IJCNN 2013. !Learning Visual Object Detection and Localisation Using icVision. Jürgen Leitner, Simon Harding, Pramod Chandrashekhariah, Mikhail Frank, Alexander Förster, Jochen Triesch, Jürgen Schmidhuber. Biologically Inspired Cognitive Architectures, Vol. 5, 2013. !Humanoid Learns to Detect Its Own Hands. Jürgen Leitner, Simon Harding, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. CEC 2013. !Cartesian Genetic Programming for Image Processing (CGP-IP). Simon Harding, Jürgen Leitner, Jürgen Schmidhuber. In Genetic Programming Theory and Practice X, pp 31-44. ISBN: 978-1-4614-6845-5. Springer, 2013.

http://Juxi.net/papers

Page 63: From Vision to Actions - Towards Adaptive & Autonomous Humanoid Robots [PhD Defense]

referencesCartesian Genetic Programming for Image Processing and Robot Vision. J. Leitner, S. Harding, A. Förster, J. Schmidhuber. IEEE RAS Summer School on "Robot Vision and Applications". Santiago, Chile. December 2012. !Learning Spatial Object Localization from Vision on a Humanoid Robot. Jürgen Leitner, Simon Harding, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. International Journal of Advanced Robotic Systems, Vol. 9, 2012. !Autonomous Learning Of Robust Visual Object Detection And Identification On A Humanoid. J. Leitner, P. Chandrashekhariah, S. Harding, M. Frank, G. Spina, A. Förster, J. Triesch, J. Schmidhuber. ICDL/EpiRob 2012. Paper of Excellence Award !An Integrated, Modular Framework for Computer Vision and Cognitive Robotics Research (icVision). Jürgen Leitner, Simon Harding, Mikhail Frank, Alexander Förster, Jürgen Schmidhuber. BICA 2012. !Transferring Spatial Perception Between Robots Operating In A Shared Workspace. J Leitner, S Harding, M Frank, A Förster, J Schmidhuber. IROS 2012. !Towards Spatial Perception: Learning to Locate Objects From Vision. J Leitner, S Harding, M Frank, A Förster, J Schmidhuber. RobotDoC-PhD 2012. !Mars Terrain Image Classification using Cartesian Genetic Programming. J Leitner, S Harding, A Förster, J Schmidhuber. i-SAIRAS 2012. !The Modular Behavioral Environment for Humanoids & other Robots (MoBeE). Mikhail Frank, Jürgen Leitner, Marijn Stollenga, Gregor Kaufmann, Simon Harding, Alexander Förster, Jürgen Schmidhuber. ICINCO 2012. !MT-CGP: Mixed Type Cartesian Genetic Programming. S. Harding, V. Graziano, J. Leitner, Jürgen Schmidhuber. GECCO 2012. !icVision: A Modular Vision System for Cognitive Robotics Research. J. Leitner, S. Harding, M. Frank, A. Förster, J. Schmidhuber. CogSys 2012. !Artificial Curiosity for Autonomous Space Exploration. Vincent Graziano, Tobias Glasmachers, Tom Schaul, Leo Pape, Giuseppe Cuccu, Jürgen Leitner and Jürgen Schmidhuber. Acta Futura, 4, pp. 41-52, 2011.

http://Juxi.net/papers2/2