Impulse-Presentation AIS / ICIAR 2010 Int. Conference on ... · s Lab o us System Autonomo Theodore...
Transcript of Impulse-Presentation AIS / ICIAR 2010 Int. Conference on ... · s Lab o us System Autonomo Theodore...
Impulse-PresentationABB Robotics, 5.12.2008
AIS / ICIAR 2010 Int. Conference on Autonomous and Intelligent Systems,Int. Conference on Autonomous and Intelligent Systems
Int. Conference on Image Analysis and RecognitionJune 21-23, 2010, Povoa de Varzim, Portugal
Roland SiegwartA t S t L bAutonomous Systems Lab
ETH Zurich
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Introduction and Challenge
Auto
nom
o
◦ “Seeing, Feeling and Understanding” the World Robots on Wheels◦ Tour-guide robotg◦ Rolling on a ball◦ Autonomous driving in cities◦ Rough terrain locomotiong
Robots on Legs◦ Quadruped running
Robots in the Air Robots in the Air◦ Micro helicopters◦ Solar airplane
Robots in the Water Robots in the Water◦ The robot tuna◦ Traversing the ocean
Conclusion
Zürich
Conclusion
s La
bou
s Sys
tem
Theodore Von Karman the famous
Auto
nom
o Theodore Von Karman, the famous aerodynamicist, once observed that:◦ Science is the study of the world as it is.
i i i h i f h ld f◦ Engineering is the creation of the world of tomorrow.
Science is basically "passive“ observation of the universe as it exists, to generate knowledge.
Engineering is making use of that knowledge to meet human needs by creating machines meet human needs by creating machines, systems, processes, and technologies that have not previously existed.
Zürich3
- Perception with multiple senses
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Reasoning about a situation Dealing with the real world
Auto
nom
o Dealing with the real world
Cognitive systems have to interpret situations based on uncertain and only partially available informationon uncertain and only partially available information
The need ways to learn functional and contextual information (models / semantics / understanding)information (models / semantics / understanding)
Probabilistic Reasoning
Zürich
s La
bou
s Sys
tem
Perception and models (“understanding”) are strongly linked
Auto
nom
o strongly linked
What is the What is the difference
in brightness?in brightness?
Zürichhttp://web.mit.edu/persci/people/adelson/checkershadow_downloads.html
s La
bou
s Sys
tem
Tactility, key for controlling the real world
Auto
nom
o
Courtesy of Albu-Schaeffer & Hirzinger, DLR, Germany
Zürich
It takes us around 14 years to learn holding a glass with an optimal force
s La
bou
s Sys
tem
Places / Situations
Auto
nom
o Places / SituationsA specific room, a meeting situation, …
Servicing / Reasoning•Functional / Contextual Relationships of Objectson
Objects
Servicing / Reasoning Relationships of Objects• imposed• learned• spatial / temporal/semanticnf
orm
atio
Doors, Humans, Coke bottle, car , …
•Models / Semantics• imposed
Interaction
• spatial / temporal/semantic
essi
ng In
FeaturesLines, Contours, Colors, Phonemes, …
• imposed• learned
Com
pre
R D t
•Models• imposedl d
Navigation
usin
g &
C
Zürich
Raw DataVision, Laser, Sound, Smell, …
• learnedFu
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Auto
nom
o
Facts and Figures (May 15 – October 20, 2002)
◦ Fully autonomous navigation and interaction in human cluttered environment environment
◦ 11 robots◦ 12 hours per day◦ 159 days of operation ◦ 159 days of operation ◦ Operational time: 13,313 hours ◦ Number of visitors: 686,000◦ Total travel distance: 3 315 km
Zürich
◦ Total travel distance: 3,315 km ◦ navigation reliability nearly 100%
s La
bou
s Sys
tem
Functional Design
Auto
nom
o g Humanoid appearance only if it is necessary for the functionality
Zürich
s La
bou
s Sys
tem
Facial expressions(Eye and eyebrow
Face tracking
LED t i
Auto
nom
o (Eye and eyebrow movements)
LED matrix
Speech synthesis
Simple speech recognition
p y
Input buttons
Obstacle avoidancePath planning
Localization
Feature extractionp g
Multi robot coordinationPeople tracking
On-board computer
Tactile sensors
Zürich
On-board computerBatteries Bumpers
s La
bou
s Sys
tem
56 s
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
Wheel design adopted from Kumagai & Ochiai, Tohoku Gakuin Universtity, Japan
s La
bou
s Sys
tem
Planar Model3D Model
Auto
nom
o
Zur Anzeige wird der QuickTime™ Dekompressor „“
benötigt.
Zürich
s La
bou
s Sys
tem
Up to 17° tilt angle Up to 3 5 m/s
Auto
nom
o Up to 3.5 m/s
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
ZürichIn collaboration with University of Freiburg
s La
bou
s Sys
tem
Motion is estimated by tracking salient points
Auto
nom
o tracking salient points The motion estimate is
sensitive to outliers se s e o ou e sthat must be removed
Immagine 1 Immagine 2
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Numb. of iterations = 1 The most efficient algorithm
f i tli for removing outliers Runs to 800 fps!
Zürich
[Scaramuzza et al., ICRA’09] [Scaramuzza et al., ICCV’09]
s La
bou
s Sys
tem
The 1-point method works optimally when the camera is on wheel axle
Auto
nom
o
However, when the cameras has an off-set something very special happens!y p pp
We can estimate the Absolute Scale from a single camera![Scaramuzza et al ICCV’09]
Zürich
[Scaramuzza et al., ICCV’09]
s La
bou
s Sys
tem Davide Scaramuzza
Auto
nom
o
ZürichE
s La
bou
s Sys
tem Davide Scaramuzza
Auto
nom
o
The video shows a 3 Km path recovered using only point p g y pfeatures and the car speed for the scale
In the real case the algorithm works even faster than this video (without feature extraction) thanks to the usage of
Zürich
video (without feature extraction) thanks to the usage of vehicle motion model
ETH-ASL, EUROPA Kick-off, 23.03.2009
s La
bou
s Sys
tem Davide Scaramuzza
Auto
nom
o
ZürichETH-ASL, EUROPA Kick-off, 23.03.2009
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
Lowenstrasse, Zurich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Passive locomotion concept 6 wheels
Auto
nom
o 6 wheels◦ two boogies on each side◦ one fixed wheel in the rear
f h l i h i ◦ one front wheel with spring suspension
length: 60 cmg height: 20 cm
Characteristics◦ highly stable in rough terrain◦ overcomes obstacles up to overcomes obstacles up to
2 times its wheel diameter
ZürichIntelligence Starts with the Design
s La
bou
s Sys
tem
Comparison of Concepts◦ CRAB (sim & HW)
Auto
nom
o ◦ CRAB (sim. & HW)◦ RCL-E (sim. & HW)◦ MER – rocker bogie type rover (sim.)g yp ( )
Zürichfront
front
s La
bou
s Sys
tem
Auto
nom
o
CRAB RCL-E MER MER
Zürich
FWD BWD
Max. [-] 0.64 0.95 0.57 1.0
Max. T [Nm] 6.0 7.3 6.7 8.9
s La
bou
s Sys
tem
Auto
nom
o
RCL-C
Crab-ETH
Zürich
s La
bou
s Sys
tem
Spirit and Opportunity Robots on Mars
Auto
nom
o Robots on Mars – since 24.1.2004
Zürich
s La
bou
s Sys
tem
Better tractive performance Lower total motion resistance
Auto
nom
o Lower total motion resistance
Courtesy of DLR Köln
Total sinkage
[mm]
Wheel deflection
[mm]
Max. soil
slope [°]
Required wheel output torque [Nm]
Combined output
power (6 wheels) [W]
Required input
power [W]
[Nm]
Rigid wheelD=35 cm, b=15 cm,
h i ht 3 4 i 10 %
45.8 - 13.9 13.87 10.6 25.2
grouser height=3.4 cm, i=10 %
Flexible wheelD=35 cm, b=15 cm, grouser height=0 1 cm pressure
12.9 12.8 13.9 6.17 4.7 11.2
Zürich
grouser height=0.1 cm, pressure on rigid ground=5 kPa, i=10 %
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Auto
nom
o
‘stiff’ robots:are built to do exactly
what they are told to what they are told to build rigid with strong
structural dampingp gand using high-gain position control
d i d suppress undesired dynamics
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
Auto
nom
o
‘stiff’ robots: might kill people i ht kill b t might kill robots kill energy
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
‘soft’ robots:
Auto
nom
o soft robots:are robust against
collisions & impactscollisions & impactscan better handle
uncertaintiescan temporarily
store energy
McGuigan & Wilson, 2003 – J. Exp. Bio.
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
Auto
nom
o
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
‘soft’ robots:
Auto
nom
o soft robots:are robust against
collisions & impactscollisions & impactscan better handle
uncertaintiescan temporarily
store energyR d k Reduce peak power
McGuigan & Wilson, 2003 – J. Exp. Bio.
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
Auto
nom
o
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
Auto
nom
o
Zürich
s La
b
Introduction Method Results Conclusion
ous
Sys
tem
The perfect actuator…p o ides high po e /to q e at a
Auto
nom
o ◦ provides high power/torque at a low weight
◦ is (sometimes) back drivable is (sometimes) back drivable ◦ has a very low reflected inertia◦ can recover negative work◦ is efficient and easy to control◦ …
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Auto
nom
o
Zürich
s La
bou
s Sys
tem
Simulation and reality
Auto
nom
o
Zürich
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Access to environments where no human or other vehicles gets
Auto
nom
o no human or other vehicles gets access to
Reducing the risk for the environment
Zürich
s La
b
Wing loading [N/m2]
ous
Sys
tem
All flying objects follows the square-cube law of scaling:
Auto
nom
o q g
3W L mg b W: weight2S b
/W S b1
S: wing area
ht[N
]
1
31/W S k W
Wei
gh
Possible Wing loading as function of weight
1/ 3/ 47W S W
Zürich
Tennekes H (1996) The Simple Science of Flight, From Insects to Jumbo Jets, MIT Press, Cambridge
s La
bou
s Sys
tem
Smaller and smaller◦ Innovative control
Auto
nom
o Innovative control approaches
◦ Design optimization◦ System leval design and ◦ System leval design and
integration
OS4 CoaX muFly
Zürich
70 cm650 g
10 cm50 g
30 cm200 g
s La
bou
s Sys
tem
General visual control framework:
Auto
nom
o
1) Start Phase 2) Init Phase 3) Flight Phase
Take off(without map)
Initializationof vSLAM framework
Navigation using vSLAM(without map) framework
(baseline & scale)
artificial l d k
Zürich
landmark
s La
bou
s Sys
tem
vSLAM algorithm (G. Klein [ISMAR 2007]):
Auto
nom
o
Tracking and mapping in separate threads
Use key frames instead of every camera frame to build map
Small and static environments
State of the art: Running in real time on a PC dual core 2GHz
Zürich
gWin/Mac/Linux), open source
s La
bou
s Sys
tem
Feature based visual 3D mapping and autonomous navigation
Auto
nom
o autonomous navigation vSLAM with Fast Corners
Zürich
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem
Develop & realize an autonomous solar powered micro glider
Auto
nom
o autonomous, solar powered micro-glider◦ Power autonomy for staying in air for days◦ Navigational autonomy◦ Fly on Earth in Martian condition (high
altit de)altitude)
Atmospheric Density ◦ ~1/80 compared with earth1/80 compared with earth
Gravity◦ ~1/3 compared with earth ?
Solar Energy◦ ~1/2 compared with earth
? Targeted Payload◦ 0.5 Kg◦ Lightweight sensors and scientific instruments
Zürich
g g◦ Atmosphere, magnetic field study
s La
bou
s Sys
tem
Based on Mass & Power Balance◦ Need for precise scaling laws
( d l )
Auto
nom
o (mass models)
Airplane PartsS l ll• Solar cells
• Battery• Airframe Aerodynamic & Conditions• …
Total mass
Aerodynamic & Conditions Power for level Flight
Zürich
s La
bou
s Sys
tem
Influence of battery technology on flight altitude on Earth
Auto
nom
o
3.2 m
Zürich
s La
bou
s Sys
tem
Continuous flight successfully demonstration on
Auto
nom
o Continuous flight successfully demonstration on
June 20 to 21, 2008 - 27 hours flight
Main Characteristics◦ 3 2 m wing span◦ 3.2 m wing span◦ 2.4 kg total weight◦ 1.2 kg of battery
P 12W ( itho t pa load)◦ PLeveled ~ 12W (without payload)◦ Very stable, even at high speeds◦ Maximum power point 91-92 % efficiency 7 grams
Zürich
ZürichAutonomous Systems Lab
s La
bou
s Sys
tem www.naro.ethz.ch
Technical Details◦ Design inspired by Tuna
Auto
nom
o ◦ Design inspired by Tuna◦ Length: 90 cm◦ Weight: ~12 kgg g◦ Maximum Speed: 2 m/s◦ Diving depth: 5 m
6 Segments, 5 actuated joints
Zürich
s La
bou
s Sys
tem
Skin
Auto
nom
o
Zürich18
s La
bou
s Sys
tem
Crossing the Atlantic
www.ssa.ethz.ch
Auto
nom
o C oss g t e t a t c◦ 4‘200 nautical Miles◦ Fully autonomous
Technical Details Technical Details◦ Very innovative design of rig◦ Length: 4m◦ Width: 1.6m◦ Over all height: 8.5m◦ Draught: 2mg◦ Weight: 530kg◦ Solar power and fuel-cells
Investment Investment◦ 15’000 Person hours◦ CHF 128‘000 cash sponsoring
CHF 70‘000 t i l i
Zürich
◦ CHF 70‘000 material sponsoring
s La
bou
s Sys
tem
Entertainment Industrial Transportation
Auto
nom
o Industrial Transportation Cleaning Medical robotics Office logistics Office logistics
Driver Support Systems The coffee servantNesspresso / Bluebotics Switzerland
Construction, mining Farming Rescuing, fire fighting, surveillance
Nesspresso / Bluebotics, Switzerland
Rescuing, fire fighting, surveillance Industrial services
Health and elderly care Services in private and public places
ZürichService Robot
ETH President greeting ASIMOV, Honda Inc.
s La
bou
s Sys
tem
“Seeing, Feeling and Understanding” the world are key issues in robotics
Places / Situations
R i
Auto
nom
o the world are key issues in robotics
Models that reflect functionalities of bj t till i i
Objects
Models / SemanticsInteraction
Reasoning Understanding
objects are still missing
Intelligence starts with the design Raw Data
Features
ModelsNavigation
Intelligence starts with the designof the most appropriate system
d d i h dl b Bad designs can hardly be compensated by control and intelligence
"One should keep things as simple as possible but not simpler!"
Zürich
possible - but not simpler! (A. Einstein)
s La
bou
s Sys
tem
Auto
nom
o
WWW: www.asl.ethz.ch
YouTube Channel: YouTube Channel: http://www.youtube.com/profile?user=aslteam&view=videos
Zürich