Docstoc

Lecture 20 Robocup

Document Sample
Lecture 20 Robocup Powered By Docstoc
					         RoboCup:
The Robot World Cup Initiative

      Based on Wikipedia and presentations by
       Mariya Miteva, Kevin Lam, Paul Marlow
What is RoboCup?

•   RoboCup is an international robotics
    competition
•   The official goal of the project
     – By mid-21st century, a team of fully
        autonomous humanoid robot soccer
        players shall win the soccer game,
        complying with the official rule of the
        FIFA, against the winner of the most
        recent World Cup.
•   Sample Game Video
                     The Different Competitions
•   RoboCup Soccer
     – Small size
     – Middle size
     – Four-legged
     – Humanoid
     – Simulation
•   RoboCup Rescue
•   RoboCup Junior
     – Soccer Challenge
     – Dance Challenge
     – Rescue Challenge
     – Genereal
Examples of RoboCup
             Leagues
     four-legged




     small size




     middle size




     simulation
RoboCup as a Standard AI
                Problem



 •   Standard problems are the driving force of AI research.
     For example, research on chess lead to the discovery of
     powerful search algorithms.
 •   AI research should be focused on solving real life
     problems, but often face social or economic constraints.
 •   RoboCup is designed to meet the need of handling real
     world complexity, though in a limited environment, while
     maintaining an affordable problem size and research cost.
                                         Why is soccer a good
                                                      option?

                                          •   Soccer challenges
                Chess      RoboCup              – dynamic environment
                                                – real-time decision making and action
Environment Static         Dynamic              – high level of uncertainty and incomplete
                                                   information
State Change Turn taking   Real time
                                                – sensor-acquired information
Information                                     – distributed control and cooperation
                Complete   Incomplete
accessibility                             •   Areas of research include real-time sensor
                                              fusion, reactive behavior, strategy acquisition,
Sensor                     Non-
                Symbolic                      learning, real-time planning, multi-agent
Readings                   symbolic
                                              systems, context recognition, vision, strategic
Control         Central    Distributed        decision making, motor control, intelligent
                                              robot control, etc.
                                                   Rules

•   There are real robot, special skill, and
    simulation competitions, each having
    different rules usually controlled by
    human referees.
•   In real robot competitions attributes of
    the environment such as the size of the
    field and the goal, the colors of the field,
    balls and robots, the maximum number
    of robots in a team, etc. are
    predetermined and differ from league to
    league.
•   Most physical fouls are considered
    unintentional and ignored.
                 More Rules


•   In simulation RoboCup a Soccer Server
    provides the virtual environment and
    controls the communication between the
    virtual robots and their control programs.
•   Robots do not know their exact position,
    but only their position relative to
    landmarks
•   Simulation allows development of
    advanced coordination systems without
    the physical constraints of real robots
    Research Issues:

•   The goal of the competitions is to stimulate
    research and advancement in both designing
    and programming robots.
•   The major areas of interest according to the
    article are:
     – Collaboration in a multi-agent
         environment
     – Design and control
     – Vision and sensor fusion
     – Learning
                                               Collaboration
•   Each team has
      – common goal (to win the game) , incompatible with the goal of the
         opponent team, and several subgoals (scoring)
      – team-wide strategies to fulfill the common goal and local and global
         tactics to achieve subgoals
•   Complications:
      – Dynamic environment
      – Locally limited perception
      – Different roles of team players
      – Limited communication among players
•   Trade-off between communication cost and accuracy of the global plan
•   Final goal - promising local plans at each agent and coordination of these
    local plans
    RoboCup Simulator




Server
Monitor clients
Player clients (i.e. agents!)
Coach clients
Sample Team Strategy
Design and Control

•   Existing robots have been designed to
    perform mostly single behavior actions
•   A RoboCup player needs to perform multiple
    subtasks( shooting, passing, heading,
    throwing, etc.) and meanwhile avoid
    opponents
•   Two approaches in building a RoboCup
    player:
     – A combination of many specialized
       components
     – One or two multitasking components
•   The final goal of building a successful
    Humanoid Soccer Player currently appears to
    be unfeasible
     Vision and Sensor Fusion


•   Computer Vision researchers have been seeking for 3D
    reconstruction of 2D visual information
•   3D reconstruction is too time-consuming for a RoboCup player to
    react in real time,
•   Other sensors (sonar, touch and force) need to be incorporated to
    provide further information, which can not be acquired by vision
•   A method of sensor fusion/integration is necessary
•   Robot Vision Video
•   Robot Vision Video 2
                                                      Learning


•   Because of the dynamic and uncertain RoboCup
    environment, programming robot behaviors for
    all possible situations is impossible.
•   Reinforcement learning is promising in
    RoboCup, since it allows acquisition of
    advanced behaviors with little prior knowledge.
•   Almost all existing reinforcement learning has
    been used in computer simulation, but not in
    physical applications.
                                                               Learning

  •   Robots first learn skills in one to one competitions.
  •   To simplify the process, task decomposition is implemented
       – two skills are independently acquired and then coordinated through learning
• Later on, many-to-many competitions
are considered.
      – It’s hard to find a simple method for
      learning collective behavior
      – Pattern finding methods or
      “coordination by imitation” are used
• The most difficult task is integration of
the learning methods in a physical
environment.
Current Champions 2008
            (Germany)
   •   Soccer
        – Small Sized: Fu-Fighters (German)
        – Middle Sized: EIGEN Keio Univ (Japan)
        – 4 Legged: German Team (German)
        – Humanoid 2-on-2: Team Osaka (Japan)
        – 2D Simulation: Brainstormers 2D (Germany)
        – 3D Simulation: Aria (Iran)
   •   Rescue
        – Simulation: Impossibles (Iran)
        – Robot: Toin Pelican (Japan)

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:4/3/2013
language:English
pages:17