RoboCup A Challenge Problem for Artificial Intelligence
Michail G. Lagoudakis Intelligent Systems Laboratory Department of Electronic and Computer Engineering Technical University of Crete Chania, Greece
Democritus Summer School 2010 July 2010, Athens, Greece RoboCup: A Challenge Problem for Artificial Intelligence
Sunday 17 July 2050 Spain vs Robots Soccer City Stadium
Summer School 2010 Michail G. Lagoudakis Page 2 RoboCup: A Challenge Problem for Artificial Intelligence Humans vs. Robots 2010
Summer School 2010 Michail G. Lagoudakis Page 3 RoboCup: A Challenge Problem for Artificial Intelligence Where do we stand today?
RoboCup 2010 Humanoid KidSize League Final (Dribblers vs. Fumanoids) Summer School 2010 Michail G. Lagoudakis Page 4 RoboCup: A Challenge Problem for Artificial Intelligence Where do we stand today?
RoboCup 2010 Humanoid TeenSize League Final (Nimbro vs. CIT Brains) Summer School 2010 Michail G. Lagoudakis Page 5 RoboCup: A Challenge Problem for Artificial Intelligence Talk Outline
RoboCup
Standard Platform League
Team Kouretes
Kouretes Research
Summer School 2010 Michail G. Lagoudakis Page 6 RoboCup
Robot Champions! RoboCup: A Challenge Problem for Artificial Intelligence RoboCup
RoboCup – international robotic soccer world cup – 1994: idea conceived by Hiroaki Kitano – today: RoboCup federation [ www.robocup.org ] Vision – “By the year 2050, to develop a team of fully autonomous humanoid robots that can win against the human world soccer champions ” – ambitious endeavor similar to sending a man to the moon – "One small step for a ROBOT, one giant leap for mankind." TM Extensions – RoboRescue: search and rescue missions – RoboCup Junior, RoboCup@home, RoboDance Summer School 2010 Michail G. Lagoudakis Page 8 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup Divisions
Summer School 2010 Michail G. Lagoudakis Page 9 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup Soccer Leagues Simulation League – 11 vs. 11 independent software agents
Small Size Robot League – 5 vs. 5 small robots up to 18 cm in diameter
Middle Size Robot League – 4 vs. 4 robots up to 50 cm in diameter
Standard Platform League – 3 vs. 3 Aldebaran Nao robots – Past: 5 vs. 5 Sony Aibo robots Humanoid Robot League – humanoid robots (kid size, teen size, adult) Summer School 2010 Michail G. Lagoudakis Page 10 RoboCup: A Challenge Problem for Artificial Intelligence Why?
Landmark project – a project with a very ambitious goal – requires a series of technological breakthrough Examples – it took about 50 years … – … from the first flight of the Wright brothers … – … to the landing of the man on the moon – it took about 50 years … – … from the invention of the first digital computer … – … to win the human world champion in Chess Question – will robots be able to play soccer like Diego Forlan in 50 years? Summer School 2010 Michail G. Lagoudakis Page 11 RoboCup: A Challenge Problem for Artificial Intelligence More Questions Why humanoids? – the world is largely engineered for humans – humanoid robots can share the same environment … – … but, they first need to acquire the same skills! – huge social and industrial impact Why soccer? – everybody in the world knows the problem/task! – it integrates all areas of Artificial Intelligence
Chess RoboCup – fully observable – partially observable – causal, static, discrete time – stochastic, dynamic, continuous time – single agent, alternating – multi agent, simultaneous Summer School 2010 Michail G. Lagoudakis Page 12 RoboCup: A Challenge Problem for Artificial Intelligence The Challenges of RoboCup What skills do I need to play soccer? – can I see clearly? [ machine vision ] – what do I see and where? [ object and event recognition ] – where am I right now? [ self localization ] – what do I do next? [ decision making and planning ] – how can I help my teammates? [ multi agent coordination ] – how can I move effectively? [ motion control ] – am I sure about what is going on? [ uncertainty ] – do I have to do everything? [ integrated solution ] – can I take my time? [ real time system ] Applicability – autonomous teams of robots in dynamic environments – search and rescue, planetary exploration, surveillance, ... Summer School 2010 Michail G. Lagoudakis Page 13 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup Workflow
[Kitano and Asada, 1998]
Summer School 2010 Michail G. Lagoudakis Page 14 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup Participation Event Place Teams Countries RoboCup 2010 Singapore, Singapore 400 40 RoboCup 2009 Graz, Austria 350 39 RoboCup 2008 Suzhou, China 397 41 RoboCup 2007 Atlanta, USA 321 33 RoboCup 2006 Bremen, Germany 440 35 RoboCup 2005 Osaka, Japan 330 31 RoboCup 2004 Lisbon, Portugal 345 37 RoboCup 2003 Padua, Italy 238 35 RoboCup 2002 Fukuoka/Busan, Japan 188 29 RoboCup 2001 Seattle, USA 141 22 RoboCup 2000 Melbourne, Australia 110 19 RoboCup 1999 Stockholm, Sweden 85 23 RoboCup 1998 Paris, France 63 19 RoboCup 1997 Nagoya, Japan 38 11 Summer School 2010 Michail G. Lagoudakis Page 15 RoboCup: A Challenge Problem for Artificial Intelligence Local RoboCup Events
Japan Open – since 1998 German Open – since 2001 US Open – since 2003 Med Open – since 2010 Other Opens – Iran, China, Brazilian, Latin American, Dutch, Spanish, ...
Summer School 2010 Michail G. Lagoudakis Page 16 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup Symposium
Symposium – main forum for presenting and discussing scientific contributions – during or after the competition – invited talks – oral and poster presentations Proceedings – Springer LNAI series – Robot Soccer World Cup xxx e Proceedings – team description papers Summer School 2010 Michail G. Lagoudakis Page 17 RoboCup: A Challenge Problem for Artificial Intelligence RoboCup and Industry
Technology transfer – develop and test in RoboCup – apply to real world problems From the fields of RoboCup … – middle size league – Minho team from Portugal – omni directional locomotion … to a commercial product – omni directional wheelchair – SAR company, Portugal – great maneuverability
Summer School 2010 Michail G. Lagoudakis Page 18 The Aldebaran Nao Robot
Your humanoid companion! RoboCup: A Challenge Problem for Artificial Intelligence Hello! My name is Nao!
Summer School 2010 Michail G. Lagoudakis Page 20 RoboCup: A Challenge Problem for Artificial Intelligence The Story of Nao
Concept – low cost robot for education, research, and entertainment – a replacement robot platform for the Sony Aibo robot Manufacturer – Aldebaran Robotics, Paris, France – small company with a team of 80 people – founder and CEO: Bruno Maisonnier Models – special limited edition for RoboCup (v.1, v.2, v.3, v.3+) – academics edition in the market since 2009 at ~10,000. € – full market release in 2010 at ~12,000. € Summer School 2010 Michail G. Lagoudakis Page 21 RoboCup: A Challenge Problem for Artificial Intelligence The Evolution of Nao
AL-01/02/03 2005 AL-04 2006 AL-05.a 2007 AL-05.b 2007 RoboCup V2 2008 RoboCup V3 2009 RoboCup V3+ 2010
Summer School 2010 Michail G. Lagoudakis Page 22 RoboCup: A Challenge Problem for Artificial Intelligence Nao Parts
Summer School 2010 Michail G. Lagoudakis Page 23 RoboCup: A Challenge Problem for Artificial Intelligence Nao Hardware
Computer – x86 AMD Geode 500MHz processor – 256 MB SDRAM memory, 2 Gb Flash memory – Embedded Linux (32 bit x86 ELF) operating system – 802.11g wireless LAN and ethernet LAN – speech synthesizer Electronics – chest board for controlling devices (Arm9, 10 ms cycle) – USB and I2C links – 6 cell, Li Ion, 21.6V battery Specifications – weight: 4.3 kg, height: 58 cm Summer School 2010 Michail G. Lagoudakis Page 24 RoboCup: A Challenge Problem for Artificial Intelligence Nao Sensors
External – 2 color CCD cameras (640x480, 30 fps, 58°) – 4 ultrasonic sensors (0.15 m – 0.70 m) – pressure sensors (4 FSRs on each foot) – bumper sensors (1 on each foot) – auditory sensors (2 microphones) Proprioceptive – encoders on each joint (12 bit, 0.1 ο precision) – inertial unit (3 accelerometers, 2 gyroscopes) – battery level sensor
Summer School 2010 Michail G. Lagoudakis Page 25 RoboCup: A Challenge Problem for Artificial Intelligence Nao Actuators
Motion – head (2 degrees of freedom) – 2 arms (4 degrees of freedom) – 2 legs (5 degrees of freedom) – pelvis (1 degree of freedom) – a total of 21 degrees of freedom Light and Sound – light (RGB LEDs in the eyes, chest, feet) – light (10x2 Blue LEDs in the ears) – sound (2 speakers, 2W)
Summer School 2010 Michail G. Lagoudakis Page 26 Standard Platform League
Where software makes the difference! RoboCup: A Challenge Problem for Artificial Intelligence Standard Platform League
Summer School 2010 Michail G. Lagoudakis Page 28 RoboCup: A Challenge Problem for Artificial Intelligence Standard Platform League
Concept – all participating teams use the same robotic platform – robots with identical hardware (computer, sensors, actuators) – focus is placed exclusively on software development – promotes the development of effective algorithms Platforms – 1997 2008: Sony Aibo quadruped robots – 2008 today: Aldebaran Nao biped robots History – four legged league (4LL) [www.tzi.de/legged] – standard platform league (SPL) [www.tzi.de/spl] Summer School 2010 Michail G. Lagoudakis Page 29 RoboCup: A Challenge Problem for Artificial Intelligence 4LL Moments
German Open 2007: Impossibles (blue) vs. Kouretes (red) Summer School 2010 Michail G. Lagoudakis Page 30 RoboCup: A Challenge Problem for Artificial Intelligence SPL Moments
RoboCup 2010: best moments of SPL final Summer School 2010 Michail G. Lagoudakis Page 31 RoboCup: A Challenge Problem for Artificial Intelligence Standard Platform League 2010
Summer School 2010 Michail G. Lagoudakis Page 32 RoboCup: A Challenge Problem for Artificial Intelligence SPL Rules
Games – two halves, teams switch sides and colors – blue/magenta team defends the blue/yellow goal Penalties – 30 seconds or permanent removal from the game – player pushing, illegal defense, leaving the field, holding the ball Referees – one head referee, two assistant referees, game controller Game Controller – software for monitoring the state of the game – communicates information to all robots Summer School 2010 Michail G. Lagoudakis Page 33 RoboCup: A Challenge Problem for Artificial Intelligence SPL Research Challenges
Autonomy – self command and on board processing Motion – legged locomotion and motion skills Vision – directed sensing: camera mounted on a movable head Localization – uncertainty, limited view of landmarks, perceptual aliasing Behavior – coordination of multiple robots, dynamic role assignment
Summer School 2010 Michail G. Lagoudakis Page 34 Team Kouretes
www.kouretes.gr RoboCup: A Challenge Problem for Artificial Intelligence Team Kouretes (Κουρήτες )
Kouretes – RoboCup team of the Technical University of Crete (2006) – ancient Cretan warriors who were protecting ancient god Zeus when he was born by Rhea in a cave in the mountains of Crete Robot Names – Επι ήδης (Epimedes) – Ίδας (Idas) – Ιάσιος (Iasios) – Παιωναίος (Paioneus) – Ηρακλής (Heracles)
Summer School 2010 Michail G. Lagoudakis Page 36 RoboCup: A Challenge Problem for Artificial Intelligence Participation in Competitions
– RoboCup 2006 [ Bremen, Germany ] – German Open 2007 [ Hannover, Germany ] – RoboCup 2007 [ Atlanta, USA ] – Rome Cup 2008 [ Rome, Italy] – RoboCup 2008 [ Suzhou, China ] – German Open 2009 [ Hannover, Germany ] – Rome Cup 2009 [ Rome, Italy] – RoboCup 2009 [ Graz, Austria ] – Med Open 2010 [ Rome, Italy] – RoboCup@SETN’10 [ Athens, Greece ] – RoboCup 2010 [ Singapore ]
Summer School 2010 Michail G. Lagoudakis Page 37 RoboCup: A Challenge Problem for Artificial Intelligence Participation in Exhibitions
– Hi Tech Innovators Partenariat 2007 [ Thessaloniki ] – TechnoTown @ Villa Torlonia 2007 [ Rome, Italy ] – Hellenic Conference on AI 2008 [ Syros, Greece ] – Athens Digital Week 2008 [ Athens, Greece ] – Festival dellaCreativita2008 [ Firenze, Italy] – Hannover Messe [ Hannover, Germany ] – Athens Digital Week 2009 [ Athens, Greece ] – SchoolFest2010 [Athens, Greece] – Micropolis2010 [Athens, Greece] – Athens Digital Week 2010 [Athens, Greece]
Summer School 2010 Michail G. Lagoudakis Page 38 RoboCup: A Challenge Problem for Artificial Intelligence Distinctions
– 2nd place[ MSRS Simulation Challenge – RoboCup 2007 ] – top 8 [ Standard Platform League, Webots – RoboCup 2008 ] – 3rd place [ Standard Platform League, Nao – RoboCup 2008 ] – 1st place [ Standard Platform League, MSRS – RoboCup 2008 ] – 1st place [ First Robotic Olympic Games, Rome Cup 2009 ] – 6th place [ Standard Platform League, Webots – RoboCup 2009 ]
Summer School 2010 Michail G. Lagoudakis Page 39 RoboCup: A Challenge Problem for Artificial Intelligence Kouretes 2010 Members
Software Architecture – Alexandros Paraschos Communication – Evangelos Vazaios Object Recognition – Emmanuel Orfanoudakis Localization/Learning – Eleftherios Chatzilaris Obstacle Avoidance – Iris Kyranou Behavior Control – Nikolaos Spanoudakis
Summer School 2010 Michail G. Lagoudakis Page 40 RoboCup: A Challenge Problem for Artificial Intelligence Scientific Contributions Publications – J. Pazis and M. G. Lagoudakis, Binary Action Search for Learning Continuous-Action Control Policies , Proceedings of the 26th Intl Conf on Machine Learning (ICML), Montreal, Canada, 2009. – G. Pierris and M. G. Lagoudakis, An Interactive Tool for Designing Complex Robot Motion Patterns , Proceedings of the IEEE Intl Conf on Robotics and Automation (ICRA), Kobe, Japan, 2009. – J. Pazis and M. G. Lagoudakis, Learning Continuous-Action Control Policies , Proceedings of the IEEE Intl Symposium on Adaptive Dynamic Programming and Reinforc ement Learning, 2009. – S. Volioti and M. G. Lagoudakis, Histogram-Based Visual Object Recognition for the 2007 Four-Legged RoboCup League , Proceedings of the 5th Hellenic Conf on Artificial Intelligence, Syros, 2008. – G. Kontes and M. G. Lagoudakis, Coordinated Team Play in the RoboCup Four-Legged League , Proceedings of the 19th IEEE Intl Conf on Tools with Artificial Intelligence (ICTAI), 2007. Diploma Theses – Ευάγγελος Βαζαίος, A Distributed Cross-Platform Communication Framework for Robotic Teams, 2010 – Ελευθέριος Χατζηλάρης, Visual-Feature-based Self-Localization for Robotic Soccer , 2009 – Ανδρέας Πανάκος, Efficient Color Recognition under Varying Illumination Conditions, 2009 – Walid Σουλάκης, Reinforcement Learning for Robotic Walk , 2009 – Ευστάθιος Βαφειάς, Kouretes MSRS Nao Simulation Team , 2009 – Γεώργιος Πιερρής, Soccer Skills for Humanoid Robots , 2009 – Ιάσων Πάζης, Learning Continuous-Action Control Policies , 2008 – ιονυσία Χρόνη, Motion Control of a Bipedal Robot using Reinforcement Learning , 2008 – Σουζάννα Βολιώτη, Histogram-Based Visual Object Recognition for the Four-Legged League , 2008 – Γεώργιος Κόντες, Coordinated Team Play in the RoboCup Four-Legged League , 2007 Summer School 2010 Michail G. Lagoudakis Page 41 Κουρήτες Research
RoboCup@TUC RoboCup: A Challenge Problem for Artificial Intelligence Monad: Software Architecture
Summer School 2010 Michail G. Lagoudakis Page 43 RoboCup: A Challenge Problem for Artificial Intelligence Narukom: Communication
Summer School 2010 Michail G. Lagoudakis Page 44 RoboCup: A Challenge Problem for Artificial Intelligence Narukom: Communication
Summer School 2010 Michail G. Lagoudakis Page 45 RoboCup: A Challenge Problem for Artificial Intelligence KCC: Color Recognition
KCC tool – capturing – labeling – learning – classification – color table
Summer School 2010 Michail G. Lagoudakis Page 46 RoboCup: A Challenge Problem for Artificial Intelligence Object Recognition
Summer School 2010 Michail G. Lagoudakis Page 47 RoboCup: A Challenge Problem for Artificial Intelligence KLoc: Localization
Monte Carlo localization with an auxiliary particle filter Summer School 2010 Michail G. Lagoudakis Page 48 RoboCup: A Challenge Problem for Artificial Intelligence KLoc: Localization
True (blue) and estimated (red) path of the robot with 6 landmarks Summer School 2010 Michail G. Lagoudakis Page 49 RoboCup: A Challenge Problem for Artificial Intelligence KLoc: Localization
Summer School 2010 Michail G. Lagoudakis Page 50 RoboCup: A Challenge Problem for Artificial Intelligence Kouretes Motion Editor
Interactive Motion Design Summer School 2010 Michail G. Lagoudakis Page 51 RoboCup: A Challenge Problem for Artificial Intelligence Kouretes Motion Editor
Summer School 2010 Michail G. Lagoudakis Page 52 RoboCup: A Challenge Problem for Artificial Intelligence Skill Learning
Balancing dynamically on one leg using the FSRs and the ankle joints Summer School 2010 Michail G. Lagoudakis Page 53 RoboCup: A Challenge Problem for Artificial Intelligence Mapping
Local map – polar topology – sonar readings – odometry Map transformations
– rotation translation translation translation right left and rotation Summer School 2010 Michail G. Lagoudakis Page 54 RoboCup: A Challenge Problem for Artificial Intelligence Obstacle Avoidance
Summer School 2010 Michail G. Lagoudakis Page 55 RoboCup: A Challenge Problem for Artificial Intelligence Behavior
Agents Systems Engineering Methodology (ASEME) Summer School 2010 Michail G. Lagoudakis Page 56 RoboCup: A Challenge Problem for Artificial Intelligence Behavior Robocup Player
get objects publish self ball self localize localize segment detect position image objects
publish select publish publish ball plan plan objects position
attack plan A go attack plan C towards c kick ball go go ball towards c towards ball lost or ballDist>0.2m goal ball can’t see ball
c published plan = C ball lost c
attack plan B go kick ball towards c pass ball ball ball lost
accept notices
Statechart for Inter and Intra Agent Control Summer School 2010 Michail G. Lagoudakis Page 57 RoboCup: A Challenge Problem for Artificial Intelligence Behavior
Statechart for Goalie Control Summer School 2010 Michail G. Lagoudakis Page 58 RoboCup: A Challenge Problem for Artificial Intelligence Putting Everything Together
RoboCup 2010: Kouretes (B) vs. CHITA (R) Summer School 2010 Michail G. Lagoudakis Page 59 RoboCup: A Challenge Problem for Artificial Intelligence Conclusion
RoboCup – a challenging problem – spans almost all areas of Artificial Intelligence – a landmark project with a very ambitious goal Participation – open to everybody (younger or older) – form a team and get started – let’s make our country a serious player Benefits – great opportunity for team work – excitement, creativity, experience
Summer School 2010 Michail G. Lagoudakis Page 60 RoboCup: A Challenge Problem for Artificial Intelligence Acknowledgements
Summer School 2010 Michail G. Lagoudakis Page 61 The End!
Thank you for your attention! Questions?