Advanced Robotics - Academia Superior...Advanced Robotics Markus Vincze Automation and Control...

Preview:

Citation preview

Advanced Robotics

Markus Vincze

Automation and Control Institute

Technische Universität Wien

vincze@acin.tuwien.ac.at

Linz, 9. Okt. 2014, IKT-Forum, Wirtschaftskammer Oberösterreich

Overview

• Vision for Robotics

• Robot success Stories

• Trends?

A Robot in Every Home

• Take robot out of the box and show your favorite

objects & places

• Easy-to-use interface

• Humans link places to

objects, e.g., furniture

• Learn objects & places

• Sensors: Odometry,

Stereo, now: + Kinect

Situated Object Detection

• Interaction of robot structures perception – Driving – floor; orientation – walls; objects – on tables and shelves

• Collects scene and object knowledge during task

execution

Table Plane Pop-out

EU Projekt CogX 2008-2012

Actual scene Stereo

Affordance: plane pop-out

Basic shapes

Model tracking

Object detection in attended

region

Affordance grasp

Learn naïve

physics

IKN saliency map [submitted ICVS 2011]

Basket: Example of Attention Points

2D + 3D saliency map

Pick-up Everything

[Fischinger, ICRA 2013]

Learning Complete Object Models

Detect cubes 3D model model based tracker

[BLORT –

v4r.acin.

tuwien.ac.at]

More Efficient: Learn from the Web

[Wohlkinger, IROS 2012]

Object Class Detection

200 object classes: http://www.3d-net.org

Results Categorization

In Clutter: Hypothesis Verification

Run classifiers object and pose hypotheses

Set of hypotheses = global scene model

Maximise: explained points

Minimise: outliers, double assignments, unexplained

[Aldoma, ECCV 2012, ICRA 2014]

Overview

• Vision for Robotics

• Robot success Stories

• Trends?

Robot Success Stories

Google Self-driving car project

● DARPA grand challenges starting 2004

● Sebastian Thrun (Stanford), Chris Urmson (CMU, Google)

● Fleet of self-driving cars, >160000 miles on public roads

in California & Nevada, San Francisco & Los Angeles

Google Driverless Car

• Based on accurate maps, GPS, odometry,

laser and radar (no cameras)

[E. Guizzo: How Google's Self-Driving Car Works, 2011]

Assisted Driving

[Dariu M. Gavrila, Mercedes E- und S-Class; ICVS 2013]

Assisted Driving

Machine Vision in Robot Inspection

Part position

Print quality

Part measuring

Pose estimation

Plan generation

Pick from box

(simple parts)

Novel Robot Application

Veebot, Mountain View, USA

Robotic Success Stories

Roomba and colleagues

● Braitenberg vehicles with a purpose

● Cheap and very simple

Robot toys, e.g. AIBO, Pleo, & therapy Paro

● More complex, programmable

● Interesting for early adopters and technophiles

irobot

Samsung

Paro

Pleo

Aibo

Romo – Toy Robot

14x11cm, 2 track motion

Face detection

Remote control

Curiosity – look for faces

Care or get bored

Tele-operate motion

Simple SDK: program robot HW

Robot Success Stories

Big Dog

● Marc Raibert (MIT, Boston Dynamics)

● Work starting mid eighties: behaviour-based

adaptation

Big Dog 2010 Quadruped 1987

DARPA Robotics Challenge

● “Help emergency personnel respond to

natural and man-made disasters”

Atlas (Boston Dynamics)

DRC - Winner

• Power of actuators: liquid-cooled motor and

driver module (Schaft, J)

• Google purchased Schaft (and many more)

Developmental Robotics - iCub

3.5 years old, 53 DOF

Platform to develop functions

more and more complex as

in children, e.g.,

Grasping/placing object

Stack objects

Motion of objects

Physical simulation of world

Learning relation to actions

Soft Robotics Inspired by human

compliance

Two „muscle“

concept

Yet difficult to control because of

compliance and flexibility

Passive dynamic walker: uses

potential energy to walk

Community and Cloud

● Willow Garage's PR2 robot and ROS

● Large repertoire of specific purpose components

● Allows rapid prototyping to plug together

application

scenarios

● Cognitive or

good

engineering?

Fetching beer video (1 week

hackathon) [Willow Garage]

Baxter – Two Arm Assembly

HOBBIT – The Mutual Care Robot

Fall Prevention and Acceptance Demographic challenge

Increasing age, highest risk: fall

50% hospital visit persons over 65

175M€ operations; 6% health costs

Robot for fall prevention/detection

Clean up floor, free paths at home

Socially connected, activity, entertainment

49 test persons in A, S, GR

70-88 Jahre, living alone, moderate impairments

Very high acceptance 87%

Rent for their home 77%

3-weeks Study in flats of older persons Haus der Barmherzigkeit

Trends?

• Rescue, hazarduous environments

Where humans cannot go, e.g., taurob

• Service robots: slow steady grow

• Soft robotics

Safe next to humans, e.g., FerRobotics

• Home robots

Google? Cloud robotics: Privacy?

Demographic challenge: Korea!

Community driven: open skill platform

• Industry: human robot collaboration

Recommended