This paper presents the map evaluation methodology developed for the Virtual Robots Rescue competition held as part of RoboCup. The procedure aims to evaluate the quality of maps produced by multi-robot systems with respect to a number of factors, including usability, exploration, annotation and other aspects relevant to robots and first responders. In addition to the design choices, we illustrate practical examples of maps and scores coming from the latest RoboCup contest, outlining strengths and weaknesses of our modus operandi. We also show how a benchmarking methodology developed for a simulation testbed effortlessly and faithfully transfers to maps built by a real robot. A number of conclusions may be derived from the experience reported in this paper and a thorough discussion is offered.

Abstract-We consider the problem of team-based robot mapping and localization using wireless signals broadcast from access points embedded in today's urban environments. We map and localize in an unknown environment, where the access points' locations are unspecified and for which training data is a priori unavailable. Our approach is based on an heterogeneous method combining robots with different sensor payloads. The algorithmic design assumes the ability of producing a map in real-time from a sensor-full robot that can quickly be shared by sensor-deprived robot team members. More specifically, we cast WiFi localization as classification and regression problems that we subsequently solve using machine learning techniques. In order to produce a robust system, we take advantage of the spatial and temporal information inherent in robot motion by running Monte Carlo Localization on top of our regression algorithm, greatly improving its effectiveness. A significant amount of experiments are performed and presented to prove the accuracy, effectiveness, and practicality of the algorithm. I. INTRODUCTIONAs a result of the evident necessity for robots to localize and map unknown environments, a tremendous amount of research has focused on implementing these primordial abilities. Localization problems have been extensively studied and a variety of solutions have been proposed, each assuming different sensors, robotic platforms, and scenarios. The increasingly popular trend of employing low-cost multi-robot teams [14], as opposed to a single expensive robot, provides additional constraints and challenges that have received less attention. A tradeoff naturally arises, because reducing the number of sensors will effectively decrease the robots' price while making the localization problem more challenging. We anticipate that team-based robots will require WiFi technology to exchange information between each other. We also foresee robots will continue to supply rough estimations of local movements, via odometry or similar inexpensive low accuracy sensors. These team-based robots have the advantage of being very affordable. It is clear, however, that these robots would not be practical in unknown environments due to their lack of perception abilities and, as such, we embrace an heterogeneous setup pairing a lot of these simple robots with a single robot capable of mapping an environment by traditional means (e.g., SLAM using a laser range finder or other sophisticated proximity sensors). Within this scenario, our goal is to produce a map of an unknown environment in real-time using the more capable robot, so that the less sophisticated robots can localize themselves.Given the sensory constraints imposed on the robots, we exploit wireless signals from Access Points (APs) that have

Abstr act From a theoretical perspective, one may easily argue (as we will in this chapter) that simulation accelerates the algorithm development cycle. However, in practice many in the robotics development community share the sentiment that "Simulation is doomed to succeed" [Brooks] p. 209. This comes in large part from the fact that many simulation systems are brittle; they do a fair-to-good job of simulating the expected, and fail to simulate the unexpected. It is the authors' belief that a simulation system is only as good as its models, and that deficiencies in these models lead to the majority of these failures. This chapter will attempt to address these deficiencies by presenting a systematic methodology with examples for the development of both simulated mobility models and sensor models for use with one of today's leading simulation engines. Techniques for using simulation for algorithm development leading to real-robot implementation will be presented, as well as opportunities for involvement in international robotics competitions based on these techniques.

A learning method capable of empowering a robot to successfully grasp a novel object through vision has recently been demonstrated, and generated much interest in the robotics community. In this paper we carefully analyze this new approach and apply dimensionality reduction techniques to decrease the number of features that need to be computed in order to classify whether a given pixel in an image is associated with a good or bad grasping point. Exploiting the ideas behind principal component analysis, we formulate two hypotheses about possible ways to eliminate certain features from training and classification. We then experimentally verify that the feature reduction significantly improves speed while retaining classification accuracy. Overall, the combination of the two hypotheses leads to a speedup factor of almost ten. The hypotheses are validated on third party synthetic data and also demonstrated on a seven degrees-of-freedom manipulator.

Abstract-While unimanual regrasping has been studied extensively, either by regrasping in-hand or by placing the object on a surface, bimanual regrasping has seen little attention. The recent popularity of simple end-effectors and dual-manipulator platforms makes bimanual regrasping an important behavior for service robots to possess. We solve the challenge of bimanual regrasping by casting it as an optimization problem, where the objective is to minimize execution time. The optimization problem is supplemented by image processing and a unimanual grasping algorithm based on machine learning that jointly identify two good grasping points on the object and the proper orientations for each end-effector. The optimization algorithm exploits this data by finding the proper regrasp location and orientation to minimize execution time. Influenced by human bimanual manipulation, the algorithm only requires a single stereo image as input. The efficacy of the method we propose is demonstrated on a dual manipulator torso equipped with Barrett WAM arms and Barrett Hands.