This paper presents the software framework established to facilitate cloud-hosted robot simulation. The framework addresses the challenges associated with conducting a task-oriented and real-time robot competition, the Defense Advanced Research Projects Agency (DARPA) Virtual Robotics Challenge (VRC), designed to mimic reality. The core of the framework is the Gazebo simulator, a platform to simulate robots, objects, and environments, as well as the enhancements made for the VRC to maintain a high fidelity simulation using a high degree of freedom and multisensor robot. The other major component used is the CloudSim tool, designed to enhance the automation of robotics simulation using existing cloud technologies. The results from the VRC and a discussion are also detailed in this work.

The U.S. Defense Advanced Research Projects Agency's (DARPA) Neovision2 program aims to develop artificial vision systems based on the design principles employed by mammalian vision systems. Three such algorithms are briefly described in this paper. These neuromorphic-vision systems' performance in detecting objects in video was measured using a set of annotated clips. This paper describes the results of these evaluations including the data domains, metrics, methodologies, performance over a range of operating points and a comparison with computer vision based baseline algorithms.

We present the formulation and implementation of an image stabilization system capable of stabilizing video with very large displacements between frames. A coarse-to-fine technique is applied in resolution and in model spaces. The registration algorithm uses phase correlation to obtain an initial estimate for translation between images; then Levenberg-Marquardt method for nonlinear optimization is applied to refine the solution. Registration is performed in color space, using a subset of the pixels selected by a gradient-based sub-sampling criterion. This software implementation runs at 5 Hz on non-dedicated hardware (Silicon Graphics R10000 workstation).

A solid-state inertial measurement unit (IMU) referred to as the MotionPak/sup TM/ is described. The IMU developed by BEI Sensors & Systems Company, consists of three micromachined quartz rate sensors and three accelerometers to provide a six degree of freedom (DOF), miniature, high reliability, low cost motion sensor which Carnegie Mellon University is utilizing to adapt to field robotic applications. Theory of operation and applications of the IMU are presented together with preliminary results in fusing the IMU data with a compass for a field robotics application.

Light power is affected when it crosses the atmosphere; there is a simple, albeit non-linear, relationship between the radiance of an image at any given wavelength and the distance between object and viewer. This phenomenon is called atmospheric scattering and has been extensively studied by physicists and meteorologists. We present the first analysis of this phenomenon from an image understanding perspective: we investigate a group of techniques for extraction of depth cues solely from the analysis of atmospheric scattering effects in images. Depth from scattering techniques are discussed for indoor and outdoor environments, and experimental tests with real images are presented. We have found that depth cues in outdoor scenes can be recovered with surprising accuracy and can be used as an additional information source for autonomous vehicles.

This paper presents a laser-based hazard detection system for rovers traveling over natural terrain. This system, which complements an existing stereo vision system, is based on a line-scanning laser ranger viewing the area just in front of the rover. This configuration of the laser yields a fast response time and very high precision, but due to the short look-ahead only an emergency stop can be commanded. The system consists of a data acquisition module and a hazard detection module. The acquisition module retrieves and validates range data, performing integrity checks of the laser system and correcting minor problems if necessary. The detection module evaluates the elevation profile of the terrain looking for evidence of three different hazards: step, ditch, and belly. The laser system has been tested in lunar-like terrain with excellent performance: zero misses and few false alarms operating at 4 Hz.

Planning the placement of stereo cameras on a mobile robot is often a balancing act between the quality of stereo data, the reliability of stereo matching, and ensuring that the cameras see enough of the robot's environment for navigation. In this paper, we present a nonlinear programming formulation to determine the optimal parameters for two stereo pairs of cameras and a camera mast for a mobile robot. We have applied this method to the design of a stereo camera mast for Nomad, a prototype rover currently under construction at Carnegie Mellon University, and present the results of designing that mast.

We present a model relating materials to the sounds they make when struck. The model exploits a shape-invariant property of materials: the angle of internal friction. We demonstrate the utility of this model in both analyzing and synthesizing sounds generated by impulsive excitation. For the analysis of impact sounds, we present an active approach for discriminating different materials by impulsively contacting (hitting) them, and sensing the interpreting the resulting sound. Experimental results suggest that shape-invariance may be encoded in the functional form of the relation between and angle of internal friction and frequency. For the recreation of impact sounds, we identify two key problems-the impact sound reconstruction problem, and the impact sound synthesis problem-and discuss preliminary considerations in developing solutions.

This paper extends the classic dead reckoning approach to the case of a mobile robot moving on uneven terrain such as the lunar surface or a construction site. In addition to the wheel encoders employed in classical odometry, the approach uses accelerometer and gyroscopic sensors, routinely used in aviation but rarely found on contemporary mobile robots. The paper derives a complementary Kalman filter that fuses accelerometer and rate gyro data to increase the accuracy of dead reckoning on uneven terrain. The paper presents empirical demonstrations on the effectiveness of the new method with mobile robot traverses over crater ground.

The paper presents a new application of computer vision to space robotics: a teleoperation interface which analyzes images sent by a mobile robot in space missions and produces position estimates based an the images. The estimates are displayed to the robot operator as additional information to prevent loss of orientation. The current version of the interface detects mountain formations in images and automatically searches for mountain peaks in a given topographic map. A new algorithm for position estimation uses a statistical description of the various disturbances and signals in the measurement process to produce estimates. The authors have tested the system with real images obtained in the Pittsburgh East and Dromedary Peak USGS quadrangles; they report significant improvements in speed and accuracy compared to previous systems.

This paper explores the possibility of using Sun altitude for localization of a robot in totally unknown territory. A set of Sun altitudes is obtained by processing a sequence of time-indexed images of the sky. Each altitude constrains the viewer to a circle on the surface of a celestial body, called the circle of equal altitude. A set of circles of equal altitude can be intersected to yield viewer position. We use this principle to obtain the position on Earth. Since altitude measurements are corrupted by noise, a least-square estimate is numerically calculated from the sequence of altitudes. The paper discusses the necessary theory for Sun-based localization, the technical issues of camera calibration and image processing, and presents preliminary results with real data.

Examines a little-studied method of manipulation-manipulation by striking an object and letting it slide. There are two parts to this problem: The inverse sliding problem, determining the velocities required to send an object to a desired configuration, and the impact problem, determining how to strike the object in order to achieve those velocities. The authors present a solution to these two problems for the class of rotationally symmetric objects and conclude with some observations about this method of manipulation.

In this paper, we describe practical, effective approaches to outdoor mapping and positioning, and present results from systems implemented for a prototype lunar rover. For mapping, we have developed a binocular head and mounted it on a motion-averaging mast. This head provides images to a normalized correlation matcher, that intelligently selects what part of the image to process (saving time), and subsamples the images (again saving time) without subsampling disparities (which would reduce accuracy). The mapping system has operated successfully during long-duration field exercises, processing streams of thousands of images. The positioning system employs encoders, inclinometers, a compass, and a turn-rate sensor to maintain the position and orientation of the rover as it traverses. The system succeeds in the face of significant sensor noise by virtue of sensor modelling, plus extensive filtering and data screening.

We address the problem of autonomously classifying objects from the sounds they make when struck, and present results from different attempts to classify various items. We extract the two most significant spikes in the frequency domain as features, and show that accurate object classification based on these features is possible. Two techniques are discussed: a minimum-distance classifier and a hybrid minimum-distance/decision-tree classifier. Results from classifier trials show that object classification using the hybrid classifier can be done as accurately as using the minimum-distance classifier, but at lower computational expense.

Reliable navigation is critical for a lunar rover, both for autonomous traverses and safeguarded remote teleoperation. This paper describes an implemented system that has autonomously driven a prototype wheeled lunar rover over a kilometer in natural, outdoor terrain. The navigation system uses stereo terrain maps to perform local obstacle avoidance, and arbitrates steering recommendations from both the user and the rover. The paper describes the system architecture, each of the major components, and the experimental results to date.

Presents a terrain mapping system for walking robots that constructs quantitative models of surface geometry, The accuracy of the constructed maps enables safe, power-efficient locomotion over the natural, rugged terrain found on planetary surfaces. The mapping system acquires range images with a laser rangefinder, preprocesses and stores the images, and constructs elevation maps from them at arbitrary resolutions, in arbitrary reference frames. To quantify performance in terms of accuracy, timing, and memory utilization, the authors conducted extensive tests in natural, rugged terrain, producing hundreds of millions of map points. The results indicate that the mapping system 1) is one of the few that can handle extremely rugged terrain, and 2) exhibits a high degree of real-world robustness due to its aggressive detection of image-based errors and in its compensation for time-varying errors.>

A surface reconstruction method is developed, based on fractal geometry, for modeling natural terrain. The method estimates dense surfaces from sparse data located in any configuration while preserving roughness. A redefinition of the temperature parameter in the stochastic regularization method is presented. It plays a critical role in controlling roughness as a function of the fractal dimension. The fractalness of surfaces reconstructed with the temperature parameter is evaluated qualitatively by applying a technique for fractal dimension estimation. As a result, it is possible to reconstruct rugged natural surfaces which preserve the original roughness from sparse data sensed by, for example, scanning laser rangefinders.>

The authors propose a method to estimate terrain roughness directly from the depth information. They estimate the fractal dimension of terrain using the fractal Brownian function approach. For experiments with real data, they extend the approach to accommodate irreguarly sampled elevation data supplied by a scanning laser rangefinder. Applying this extended method to noisy range imagery of natural terrain (sand and rocks), the authors find that the resulting estimates of fractal dimension correlate closely to human perception of the roughness of the terrain, showing that the fractal dimension of the sensed point set is a practical and effective measure of the roughness of natural terrain.

The authors quantify several performance metrics for the Ambler, a six-legged robot configured for autonomous traversal of Mars-like terrain. They present power consumption measures for walking on sandy terrain and for vertical lifts at different velocities. They document the accuracy of a novel dead reckoning approach, and analyze the accuracy. They describe the results of autonomous walking experiments in terms of terrain traversed, walking speed, number of instructions executed and endurance.>

We describe the perception system for the Ambler, an autonomous, legged mobile robot that operates in rugged environments. The Ambler's perception system uses 3-D laser range images to construct elevation maps of the local terrain, processing hundreds of images and thousands of elevation points during a walking experiment. This paper presents the design, implementation, and analysis of a perception system to meet the needs of long-duration walking, emphasizing issues of perforumnce, accuracy, and reliability.

Autonomous and teleoperated mobile robots require an accurate knowledge of their spatial location in order to accomplish many tasks. Many mobile robots make use of dead reckoning navigation because of its simplicity, low cost and robustness. Although dead reckoning navigation has been used for centuries for ships and wheeled vehicles, the application to a walking machine is novel. Since walking machines differ greatly from ships and wheeled vehicles, a new approach to dead reckoning was developed to solve this problem. This paper discusses the problem, a solution, preliminary test results and future goals for dead reckoning navigation. Experiments were done with the CMU Ambler, an autonomous, six-legged walking robot, but the results are general and apply to any statically stable walking robot. The current results show a systematic bias of two percent of body advance in the direction of travel. Although the cause of this bias is unknown, it is corrected in the position estimation routines.

The author presents techniques for calibrating scanning laser range finders to legged robots. The procedure accommodates two scanners, one manufactured by Erim and the other by Perceptron, and two walkers, a one-legged robot and a six-legged robot. The technique acquires two sets of corresponding three-dimensional points and identifies the rigid transformation that maps one onto the other with least squared error, i.e. it solves the absolute orientation problem. Experimental results are reported with the two different scanners and vehicles. For the Erim and the one-legged robot, the technique achieves an accuracy of 6-12 cm with a precision no lower than 2-5 cm. For the Perceptron and the Ambler, the accuracy is 2-7 with a precision no lower than 2-5 cm. These results have proven to be satisfactory for constructing terrain maps and using them to select footholds during rough terrain walking experiments.>

The Carnegie Mellon University Planetary Rover project is developing the Ambler, a six-legged robot designed for planetary exploration. The authors have developed reliable and efficient control, perception and planning algorithms suitable for navigating rugged terrain. The components have been integrated into a system that autonomously walks the Ambler along routes and over obstacles.>

The authors analyze a class of imaging range finders-amplitude-modulated continuous-wave laser radars-in the context of computer vision and robotics. The analysis develops measurement models from the fundamental principles of laser radar operation, and identifies the nature and cause of key problems that plague measurements from this class of sensors. They classify the problems as fundamental (e.g. related to the signal-to-noise ratio), as architectural (e.g. limited by encoding distance by angles (0.2 pi )), and as artifacts of particular hardware implementations (e.g. insufficient temperature compensation). Experimental results from two different scanning laser range finders designed for autonomous navigation illustrate and support the analysis.>

The Ambler is an autonomous orthogonal walking robot that operates in rugged environments. The mechanism has two stacks and six circulating legs. On top of the Ambler is mounted a laser scanner to obtain 3D data of the local terrain. It is designed for planetary exploration. During the course of walking experiments unexpected shortcomings in the robot's perception system were noted. Unusual perception sensor behavior, random noise sources, and interaction of the perception system with other system components caused both minor errors and large failures in the system. This paper analyzes the sources of the perception system errors, presents solutions to these problems, and proposes a design for a more robust perception system.>

A method of computing absolute range from stereo disparities by using verging cameras is presented. The approach differs from others by concentrating, through both analysis and experiment, on the effects caused by convergence, rather than on the general camera calibration problem. To compute stereo disparities, linear image features are extracted and matched using a hypothesize-and-verify method. To compute range, the relationship between object distance, vergence angle, and disparity is derived. Experimental results show the precision of the range computation, excluding mistaken matches, to be approximately 5% for object distances up to three meters and a baseline distance of 13 cm. Including mistaken matches results in performance that is an order of magnitude worse, leading the authors to suggest methods to identify and model them.>

The authors are prototyping a legged vehicle, the Ambler, for an exploratory mission on another planet, conceivably Mars, where it is to traverse uncharted areas and collect material samples. They describe how the rover can construct from range imagery a geometric terrain representation, i.e., elevation map that includes uncertainty, unknown areas, and local features. First, they present an algorithm for constructing an elevation map from a single range image. By virtue of working in spherical-polar space, the algorithm is independent of the desired map resolution and the orientation of the sensor, unlike algorithms that work in Cartesian space. Secondly, the authors present a two-stage matching technique (feature matching followed by iconic matching) that identifies the transformation T corresponding to the vehicle displacement between two viewing positions. Thirdly, to support legged locomotion over rough terrain, they describe methods for evaluating regions of the constructed elevation maps as footholds.>

Following and extending the approach of K. Sugihara (1988), the author assumes that a mobile robot is equipped with a single camera and a map marking the positions in its environment of landmarks. The robot moves on a flat surface, acquires one image, extracts vertical edges from it, and computes the directions to visible landmarks. The problem is to determine the robot's position and orientation (pose) by establishing the correspondence between landmark directions and points in the map. This approach capitalizes on the excellent angular resolution of standard CCD cameras, while avoiding the feature-correspondence and 3D reconstruction problems. The problem is formulated as a search in a tree of interpretations (pairings of landmark directions and landmark points), and an algorithm to search the tree efficiently to determine the solution poses(s) is developed, taking into account errors in the landmark directions extracted by image processing. Quantitative results from simulations and experiments with real imagery are presented.>

A cooperative computer-vision procedure is presented in which focus ranging and stereo ranging operate together, verifying the results of each other in computing the position (but not the shape) of arbitrary objects in a stationary, unknown environment. The procedure increases the reliability of the position measurements by enforcing measurement consistency by mutual constraint, and increases their accuracy by combining them with a maximum-likelihood estimator into an estimate of lower variance than any of the measurements alone. The final outcome of the procedure is a set of estimated three-dimensional points together with their estimated uncertainties built from a sequence of dynamic, adaptive sensing operations. The results of 75 experiments processing close to 3000 different object points lying between 1 and 3 m distant from the cameras show that the integrated range values are (1) highly reliable, since no mistaken combined range measurements are observed, and (2) more precise than either of the computed ranges alone. The cooperative methodology extends to more and different sensors, and the results lend practical credence to the view that multiple sensors (1) allow the limitations of a single sensor to be circumvented: (2) provide larger, statistically more effective data sets; and (3) reduce mistakes generated by inaccurate interpretation models, using sensors outside of their known operating regions, and sensor failure.>

An agile stereo camera system designed for flexible image acquisition under a wide variety of viewing conditions and scenes is presented. A host processor sends commands to three microprocessors controlling, through 11 servomotors, the position and optical parameters of the camera, and the scene illumination. The weaknesses of the system are acknowledged and future improvements are outlined.>

One method for improving the vision system of a robot is to allow the robot to automatically focus its cameras and to use focus as a depth cue. In this paper, the problem of automatic focusing is reduced, by a theoretical analysis of defocus as a wave aberration which attenuates high spatial-frequencies, to the problem of measuring the high spatial-frequency content of an image. The problem of recovering range from focus is treated using a focus criterion function along with knowledge of the lens parameters to solve the lens law for range. The accuracy of the range measurements is approximately 10% of object distances between 1 and 2m.