Microsoft Robotics Soccer Challenge: Movement Optimization...
Transcript of Microsoft Robotics Soccer Challenge: Movement Optimization...
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 11
Microsoft Robotics SoccerChallenge: Movement Optimization
of a Quadruped RobotMestrado Integrado em Engenharia Electrotécnica e de Computadores
João Oliveira
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 22
Presentation Outline
� Introduction
� Objectives� Robotic Soccer
� Optimization and Metaheuristics
� Project and Implementation
� Results and Discussion
� Conclusion and Future Work
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 33
Introduction
� What is a robot?
� One definition may be that it is a physical agent that generates intelligent connection between perception and action
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 44
Legged Locomotion
� Two main means of locomotion: legs or wheels.
� Why is legged locomotion so common among animals?
� Why isn’t it widely used in human built machines?
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 55
Legged Locomotion
� Legged locomotion requires full coordination
� Complex task to control
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 66
Objectives
� Improve the movement of a four-legged robot, using a set of optimization algorithms
� Understand which optimization algorithms can be used in the optimization of the movement of a four-legged robot
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 77
RoboCup
� International robotics project that started in 1997
� Provides a standard problem where a wide range of technologies can be integrated and examined
� Four major types of competitions: RoboCup Soccer, RoboCup Rescue, RoboCup Junior and RoboCup@Home
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 88
RoboCup
“By mid-21st century, a team of fully autonomous humanoid robot soccer players shall win the soccer game, comply with the official rule of the FIFA, against the winner of the most recent World Cup.”
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 99
Microsoft Robotics Soccer Challenge
� Started in 2007� Aims to promote the Microsoft Robotics Studio with the
potential of the RoboCup� 2 vs 2 games of simulated robuDOG robots� Simplified environment� Complexity of the robot turned this league into a big
challenge
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1010
Microsoft Robotics Soccer Challenge
robuDOG
� Developed with theRoboCup in mind
� Front legs designed to kick the ball with betteraccuracy
� Powerful 3D SimulationEnvironment
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1111
robuDOG
� Single movement:� Front right� Front left� Rear right� Rear left
� Leg movement:� Shoulder: forward and
backward� Shoulder: left and right� Knee: forward and
backward� Ankle (front legs only)
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1212
Optimization and Metaheuristics
Optimization
� From an initial model, its parameters are progressively refined in the light of experience.
� The goal is to determine the maximum or minimum value of some function of one or more variables.
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1313
Optimization and Metaheuristics
Metaheuristics
� Methods used to search the solution space with a strategy to find optimum solutions
� Incorporates concepts based on biological evolution, intelligent problem solving, mathematical and physical sciences, nervous systems and statistical mechanics
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1414
Hill-Climbing
Hill-Climbing Algorithm:
1. Generate an initial solution.2. Loop until a stop criterion is met or there no untested neighbour solutions:
(a) Test a neighbour solution to the current one that hasn’t been tested yet
(b) If the new solution is better than thecurrent solution make it the current solution
3. Return the current solution
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1515
Hill-Climbing
Weaknesses
� Local maximum� Plateau� Ridges
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1616
Tabu Search
� Allows non-improving moves� Keeps a tabu list so that the algorithm doesn’t
go back to past solutions� With this it tries to solve the weaknesses of the
Hill-Climbing Algorithm
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1717
Simulated Annealing
� Similar to the atoms rearranging themselves in cooling metal
� It can accept worse solutions with a probability dependent of the “temperature”
� It can escape local optima to find the global optimum
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1818
Project
� Slow process, and high number of iterations needed to produce results� This eliminated the Tabu Search and other
algorithms that would require more computation time
� Focus on rotation movement� Focus on reducing iteration time and noise
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 1919
Project
� Metaheuristics tested:� Hill-Climbing� A simplified version of
Tabu Search� Simulated Annealing
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2020
Project
Simplified Tabu Search
� Tries to increase speed of optimization� Changes that improved
the solution before have a higher probability of being tested
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Hill-Climbing History list
Simplified Tabu
Search
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2121
Project
� Evaluation function for the rotation movement� Maximize rotation of the robot ( )� Minimize translation of the robot (p)
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
pSf ∆−∆= 100)( θ
θ
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2222
Results
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2323
Results
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2424
Results
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2525
Results
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2626
Results
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2727
Results
� High iteration time, making it impossible or unpractical to test exhaustively for more complex movements
� Big influence of noise and variation between results, making it difficult to evaluate each set of parameters
� Improvements were nonetheless convincing
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2828
Conclusions
� Metaheuristics are effective in this problem
� Initial solution is important, with a good one even the Hill-Climbing Algorithm performs well
� Noise and simulation time are problems that affect the effectiveness of this method
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 2929
Conclusions
Possible future changes
� Find a way to make the simulator faster, optimizing it for this approach, and using a faster computer
� Testing several times each solution to make the method more immune to noise
� After these problems are addressed, other algorithms can be used, like Tabu Search or Genetic Algorithms
Introduction | Objectives | Robotic Soccer | Opt. and Met. | Project | Results | Conclusions
Microsoft Robotics Soccer Challenge: Movement Optimization of a Microsoft Robotics Soccer Challenge: Movement Optimization of a Quadruped Robot Quadruped Robot 3030
Microsoft Robotics SoccerChallenge: Movement Optimization
of a Quadruped RobotQuestions?
João [email protected]