Paul A. Vallejos

Back to reality: Crossing the reality gap in evolutionary robotics

Abstract In this work a new method to evolutionary robotics is proposed, it combines into asingle framework, learning from reality and simulations. An illusory sub-system is incorporated as an integral part of an autonomous system. The adaptation of the illusory system results from minimizing differences of robot behavior evaluations in reality and in simulations. Behavior guides the illusory adaptation by sampling task-relevant instances of the world. Thus explicit calibration is not required. We remark two attributes of the presented methodology: (i) it is a promising approach for crossing the reality-gap among simulation and reality in evolutionary robotics, and (ii) it allows to generate automatically models and theories of the real robot environment expressed as simulations. We present validation experiments on locomotive behavior acquisition for legged robots.

Motion detection and tracking for an AIBO robot using camera motion compensation and kalman filtering

Motion detection and tracking while moving is a desired ability for any soccer player. For instance, this ability allows the determination of the ball trajectory when the player is moving himself or when he is moving his head, for making or planning a soccer-play. If a robot soccer player should have a similar functionality, then it requires an algorithm for real-time movement analysis and tracking that performs well when the camera is moving. The aim of this paper is to propose such an algorithm for an AIBO robot. The proposed algorithm uses motion compensation for having a stabilized background, where the movement is detected, and Kalman Filtering for a robust tracking of the moving objects. The algorithm can be adapted for almost any kind of mobile robot. Results of the motion detection and tracking algorithm, working in real-world video sequences, are shown.

Design and validation of a fuzzy longitudinal controller based on a vehicle dynamic simulator

This paper presents a fuzzy longitudinal controller for a wide range of speeds, implementing ACC and Stop&Go control, along with the development of a vehicle dynamics simulator for testing the performance of the fuzzy controller. The fuzzy longitudinal controller has two inputs (speed error and integral of the speed error) and a unique output signal manipulating the opening of the throttle or the brake, according to the sign of the output signal. Several simulations were carried out for adjusting controller parameters and analyzing its performance. Simulation results shows that the proposed fuzzy longitudinal controller achieves excellent speed tracking in a wide range of speeds with high accuracy and minimum steady state error.

An Automated Refereeing and Analysis Tool for the Four-Legged League

The aim of this paper is to propose an automated refereeing and analysis tool for robot soccer. This computer vision based tool can be applied for diverse tasks such as: (i) automated game refereeing, (ii) computer-based analysis of the game, and derivation of game statistics, (iii) automated annotations and semantic descriptions of the game, which could be used for the automatic generation of training data for learning complex high-level behaviors, and (iv) automatic acquisition of real game data to be used in robot soccer simulators. The most attractive application of the tool is automated refereeing. In this case, the refereeing system is built using a processing unit (standard PC) and some static and/or moving video cameras. The system can interact with the robot players and with the game controller using wireless data communication, and with the human spectators and human second referees by speech synthesis mechanisms or using visual displays. We present a refereeing and analysis system for the RoboCup Four-Legged League. This system is composed by three modules: object perception, tracking, and action analysis. The camera placement issue is solved by human controlled cameras placement and movement. Some preliminary experimental results of this system are presented.

Context - dependent color segmentation for Aibo robots

Color segmentation is a key step in any color-based robot vision system. The main factors that affect the performance of the segmentation process are the variable lighting conditions of the environment, the natural overlap between the color classes in the color space, and the distortion produced in the images by the movement of the camera and the robot. In the context of the robot soccer championship RoboCup, several segmentation approaches have been proposed, most of them based on an off-line calibration phase where a look-up table is generated for the on-line system operation. The here proposed approach employs ambiguous colors as a way of including in the calibration process the natural uncertainness of the problem. Thus, the color segmentation result is not completely defined in the calibration phase. During on-line operation a mode filter is employed for incorporating context information in the discrimination of the final class of the colors. Successful results of the application of the proposed dynamic color segmentation methodology are shown in the context of the RoboCup four-legged league

Designing Fall Sequences That Minimize Robot Damage in Robot Soccer

In this paper is proposed a methodology for the analysis and design of fall sequences that minimize robot damage. This methodology minimizes joint/articulation injuries, as well as damage of valuable body parts (cameras and processing units). The methodology is validated using humanoid Nao robots and a realistic simulator. The obtained results show that fall sequences designed using the proposed methodology produce less damage than standard, uncontrolled falls.

Probabilistic kick selection in robot soccer

Kick selection is an important issue in robot soccer that has not been investigated deeply enough by the RoboCup research community. This paper proposes a probabilistic approach to kick selection, which includes kick objective selection based on priorities. The proposed selection methodology is based on the maximization of the probability of success of the kick (the ball reaches the objective position). Experimental results of the proposed methodology show a high kick success rate, and give evidence of achieving accurate objectives even when using a set of imprecise kicks

A Kalman-filtering-based Approach for Improving Terrain Mapping in off-road Autonomous Vehicles

The generation of accurate terrain maps while navigating over off-road, irregular terrains is a complex challenge, due to the difficulties in the estimation of the pose of the laser rangefinders, which is required for the proper registration of the range measurements. This paper addresses this problem. The proposed methodology uses an Extended Kalman filter to estimate in real-time the instantaneous pose of the vehicle and the laser rangefinders by considering measurements acquired from an inertial measurement unit, internal sensorial data of the vehicle and the estimated heights of the four wheels, which are obtained from the terrain map and allow determination of the vehicle’s inclination. The estimated 6D pose of the laser rangefinders is used to correctly project the laser measurements onto the terrain map. The terrain map is a 2.5D map that stores in each cell the mean value and variance of the terrain height. In each map’s cell position, new laser observations are fused with existing height estimations using a Kalman filter. The proposed methodology is validated in the real world using an autonomous vehicle. Field trials show that the use of the proposed state estimation methodology produces maps with much higher accuracy than the standard approaches.

A New Methodology for the Design of Passive Biped Robots: Determining Conditions on the Robot's Parameters for the Existence of Stable Walking Cycles

Currently, passive robots are designed following a trial and error process in which the existence of a stable walking cycle for a given passive robot’s model is analyzed using Poincaré maps. The standard stability analysis procedure suffers from discretization aliasing, and it is not able to deal with complex passive models. In this paper a methodology that allows finding conditions on the robot’s parameters of a given passive model in order to obtain a stable walking cycle is proposed. The proposed methodology overcomes the aliasing problem that arises when Poincaré sections are discretized. Basically, it implements a search process that allows finding stable subspaces in the parameters’ space (i.e., regions with parameters’ combinations that produce stable walking cycles), by simulating the robot dynamics for different parameters’ combinations. After initial conditions are randomly selected, the robot’s dynamics is modeled step by step, and in the Poincaré section the existence of a walking cycle is verified. The methodology includes the definition of a search algorithm for exploring the parameters’ space, a method for the partition of the space in hypercubes and their efficient management using proper data structures, and the use of so-called design value functions that quantify the feasibility of the resulting parameters. Among the main characteristics of the proposed methodology are being robot independent (it can be used with any passive robot model, regardless of its complexity), and robust (stable subspaces incorporate a stability margin value that deals with differences between the robot’s model and its physical realization). The methodology is validated in the design process of a complex semi-passive robot that includes trunk, knees, and non-punctual feet. The robot also considers the use of actuators, controllers and batteries for its actuation.

Motion Detection and Object Tracking for an AIBO Robot Soccer Player 1

Movement analysis is a fundamental ability for any kind of robot. It is especially important for determining and understanding the dynamics of the robot’s surrounding environment. In the case of robot soccer players, movement analysis is employed for determining the trajectory of relevant objects (ball, teammates, etc.). However, most of the existing movement analysis methods require the use of a fixed camera (no movement of the camera while analyzing the movement of objects). As an example, the popular background subtraction algorithm employs a fixed background for determining the foreground pixels by subtracting the current frame with the background model. The requirement of a fixed camera restricts the real-time analysis that a soccer player can carry out. For instance, a human soccer player very often requires the determination of the ball trajectory when he is moving himself, or when he is moving his head, for making or planning a soccer-play. If a robot soccer player should have a similar functionality, then it requires an algorithm for real-time movement analysis that can perform well when the camera is moving. The aim of this work is to propose such an algorithm for an AIBO robot. This algorithm can be adapted for almost any kind of mobile robot. The rationale behind our algorithm is to compensate in software the camera movement using the information about the robot body and robot head movements. This information is used to correctly align the current frame and the background. In this way, a stabilized background is obtained, although the camera is always moving. Afterward, different traditional movement analysis algorithms can be applied over the stabilized background. Another feature of our algorithm is the use of a Kalman Filter for the robust tracking of the moving objects. This allows to have reliable detections and to deal with common situations such as double detections or no detection in some frames because of variable lighting conditions. This chapter is organized as follows. In section 2 we present some related work. In section 3 is described the here proposed motion analysis algorithm for AIBO robots. Experiments