Catálogo de publicaciones - libros

Robot recognition is a very important point for further improvements in game-play in middle size league. In this paper we present a neural recognition method we developed to find robots using different visual information. Two algorithms are introduced to detect possible robot areas in an image and a subsequent recognition method with two combined multi-layer perceptrons is used to classify this areas regarding different features. The presented results indicate a very good overall performance of this approach.

Humans process images with apparent ease, quickly filtering out useless information and identifying objects based on their shape and colour. However, the undertaking of visual processing and the implementation of object recognition systems on a robot can be a challenging task. While many algorithms exist for machine vision, fewer have been developed with the efficiency required to allow real-time operation on a processor limited platform. This paper focuses on several efficient algorithms designed to identify field landmarks and objects found in the controlled environment of the RoboCup Four-Legged League.

In competitive domains, the knowledge about the opponent can give players a clear advantage. This idea lead us in the past to propose an approach to acquire models of opponents, based only on the observation of their input-output behavior. If opponent outputs could be accessed directly, a model can be constructed by feeding a machine learning method with traces of the opponent. However, that is not the case in the Robocup domain. To overcome this problem, in this paper we present a three phases approach to model low-level behavior of individual opponent agents. First, we build a classifier to label opponent actions based on observation. Second, our agent observes an opponent and labels its actions using the previous classifier. From these observations, a model is constructed to predict the opponent actions. Finally, the agent uses the model to anticipate opponent reactions. In this paper, we have presented a proof-of-principle of our approach, termed OMBO (Opponent Modeling Based on Observation), so that a striker agent can anticipate a goalie. Results show that scores are significantly higher using the acquired opponent’s model of actions.

Despite efforts to design precise motor controllers, robot joints do not always move exactly as desired. This paper introduces a general model-based method for improving the accuracy of joint control. First, a model that predicts the effects of joint requests is built based on empirical data. Then this model is approximately inverted to determine the control requests that will most closely lead to the desired movements. We implement and validate this approach on a popular, commercially available robot, the Sony Aibo ERS-210A.

This paper presents a new approach to optimize gait parameter sets using evolutionary algorithms. It separates the crossover-step of the evolutionary algorithm into an interpolating step and an extrapolating step, which allows for solving optimization problems with a small population, which is an essential for robotics applications. In contrast to other approaches, odometry is used to assess the quality of a gait. Thereby, omni-directional gaits can be evolved. Some experiments with the Sony Aibo models ERS-210 and ERS-7 prove the performance of the approach including the fastest gait found so far for the Aibo ERS-210.

Generation of a sequence of behaviors is necessary for the RoboCup Humanoid league to realize not simply an individual robot performance but also cooperative ones between robots. A typical example task is passing a ball between two humanoids, and the issues are: (1) basic skill decomposition, (2) skill learning, and (3) planning to connect the learned skills. This paper presents three methods for basic skill learning (trapping, approaching to, and kicking a ball) based on optic flow information by which a robot obtains sensorimotor mapping to realize the desired skill, assuming that skill decomposition and planning are given in advance. First, optic flow information of the ball is used to predict the trapping point. Next, the flow information caused by the self-motion is classified into the representative vectors, each of which is connected to motor modules and their parameters. Finally, optical flow for the environment caused by kicking motion is used to predict the ball trajectory after kicking. The experimental results are shown and discussion is given with future issues.

Kicking the ball with high power, short reaction time and accuracy are fundamental requirements for any soccer player. Human players acquire these fine low-level sensory motor coordination abilities trough extended training periods that might last for years. In RoboCup the problem has been addressed by engineering design and acceptable, probably sub-optimal, solutions have been found. To our knowledge the automatic development of these abilities has not been yet employed. Certainly no one is willing to damage a robot during an extended, and probably violent, evolutionary learning process in a real environment. In this work we present an approach for the automatic generation (from scratch) of ball-kick behaviors for legged robots. The approach relies on the use of UCHILSIM, a dynamically accurate simulator, and the paradigm to evolutionary robotics, a recently proposed method for narrowing the difference between simulation and reality during robot behavior execution. After eight hours of simulations successful ball-kick behaviors emerged, being directly transferable to the real robot.

The joints of a humanoid robot experience disturbances of markedly different magnitudes during the course of a walking gait. Consequently, simple feedback control techniques poorly track desired joint trajectories. This paper explores the addition of a control system inspired by the architecture of the cerebellum to improve system response. This system learns to compensate the changes in load that occur during a cycle of motion. The joint compensation scheme, called Trajectory Error Learning, augments the existing feedback control loop on a humanoid robot. The results from tests on the GuRoo platform show an improvement in system response for the system when augmented with the cerebellar compensator.

Humanoid dynamic walk and kick are two main technical challenges for the current Humanoid League. In this paper, we conduct a research aiming at generating dynamically stable walking and kicking gait for humanoid soccer robots with consideration of different constraints. Two methods are presented. One is synthesizing gait based on constraint equations, which has formulated gait synthesis as an optimization problem with consideration of some constraints, e.g. zero-moment point (ZMP) constraints for dynamically stable locomotion, internal forces constraints for smooth transition, geometric constraints for walking on an uneven floor and etc. The other is generating feasible gait based on human kicking motion capture data (HKMCD), which uses periodic joint motion corrections at selected joints to approximately match the desired ZMP trajectory. The effectiveness of the proposed dynamically stable gait planning approach for humanoid walking on a sloping surface and humanoid kicking on an even floor has been successfully tested on our newly developed Robo-Erectus humanoid soccer robots, which won second place in the RoboCup 2002 Humanoid Walk competition and got first place in the RoboCup 2003 Humanoid Free Performance competition.

This paper proposes a new algorithm for the automatic segmentation of motion data from a humanoid soccer playing robot that allows feed-forward neural networks to generalize and reproduce various kinematic patterns, including walking, turning, and sidestepping. Data from a 20 degree-of-freedom Fujitsu -1 robot is reduced to its intrinsic dimensionality, as determined by the procedure, by means of nonlinear principal component analysis (). The proposed algorithm then automatically segments motion patterns by incrementally generating periodic temporally-constrained nonlinear neural networks and assigning data points to these networks in a -and-divide fashion, that is, each network’s ability to learn the data influences the data’s division among the networks. The learned networks abstract five out of six types of motion without any prior information about the number or type of motion patterns. The multiple decoding subnetworks that result can serve to generate abstract actions for playing soccer and other complex tasks.