Humanoid Robot NAO: Developing Behaviors for Football Humanoid Robots State of the Art Presentation Luís Miranda Cruz Supervisors: Prof. Luis Paulo Reis Prof. Armando Sousa
Outline 1. Context 1.1. Robocup 1.2. FCPortugal 2. State of the Art 2.1. Humanoids 2.2. Simulation 2.5. Humanoid Behaviours 3. The Problem 4. Methodology 4.1. Proposed Architecture 5. Conclusion - Work Plan 6. References 2.3. Optimization 2.4. Machine Learning
Robocup Robot Soccer World Cup Challenge the research in Artificial Intelligence, Robotics and related fields.
FCPortugal RoboCup Team Cooperation between Universities of Porto and Aveiro Has won many awards in several RoboCup competitions since 2000
Humanoids A humanoid robot is a robot with a human-like appearance. RoboNova TOPIO Asimo QRIO NAO
Simulation Environment Main advantages: Less expensive than real robots; Easier and faster testing. Tests without damage of robots. Retrieve useful information from execution. There are many simulaters available RobotSim, Webots, SimSpark
SimSpark Simulator Official Robocup Simulator Agents Server Monitor Perceptors Models Efectors SoccerBot Nao
Optimization (1/2) Find the best elements from a set of possible alternatives, according to a criteria.
Optimization (2/2) Optimization Problems are defined with: Decision Parameters Restrictions Over the Decision Parameters Objective Function Methods can be classified as: Individual-based or Population-Based Online or Offline
Machine Learning (1/2) Allows a machine to learn by itself with reduced need for human intervention. Three main categories: Supervised Learning Unsupervised Learning Reinforcement Learning
Machine Learning (2/2) There are several applications: Speech Recognition Cells Classification Automated Driving Data Mining
Humanoid Behaviors Behaviors have to be developed in order to achieve autonomous control. Three main concepts: Motor level - commands expressed directly to the system s actuators. Skill level - programmed actions without reasoning about objectives or the environment. Task level - programs of skills directed towards achieving specific goals.
Task Level Behaviours Task Go to the ball and kick it. Non trivial problem: Many possible sequences of skills to execute the same task.
The Problem Humanoid players must be capable of walking in many directions kicking the ball with high precision getting up when necessary... It s necessary to classify the available skill level behaviours. to predict the best sequence of skills for a desired task, with regard to the environment. Simultaneously in simulated and real environments.
Methodology Classificate behaviours by its quality using machine learning algorithms. Getting better results as experience grows. Use optimization algorithms to predict the best behaviour sequence for a given task. SimSpark to test in simulated environment Nao robot to test in real environment.
Proposed Architecture Behaviours Knowledge Classification Engine Execution Task level Behaviours Generator
Conclusion - Work Plan
References Picado, Hugo; "Development of Behaviors for a Simulated Humanoid Robot". PhD thesis, University of Aveiro, 2008. Reis, Luís Paulo; Lau, Nuno; "Paper Contributions, Results and Work for the Simulation Community of FC Portugal". Lau, Nuno; Reis, Luis Paulo; Picado, Hugo; Almeida, Nuno; "FCPortugal: Simulated humanoid robot team description proposal for robocup 2008". In Proceedings CD of RoboCup 2008, 2008. Picado, Hugo; "FCPortugal3D A team of RoboCup 3D Simulation League". IEETA, University of Aveiro, 2009. Dominey, Peter Ford; Lallee, Stephane; Khamassi, Mehdi; Lu, Zhenli; Lallier, Corentin; Boucher, Jean-David; Weitzenfeld, Alfredo; Ramos, Carlos; "Cooperative Human Robot Interaction with the Nao Humanoid: Technical Description Paper for the Radical Dudes" Thomas Weise. Global Optimization Algorithms - Theory and Application. 2nd edition, 2009.
Humanoid Robot NAO: Developing Behaviors for Football Humanoid Robots State of the Art Presentation Luís Miranda Cruz Supervisors: Prof. Luis Paulo Reis Prof. Armando Sousa