In this work, we have proposed the use of flow not induced by motion but by active manipulation of the lens focal length. The sequence of images thus obtained are to a great extent equivalent to those generated by camera translation with several distinct advantages: (1) the platform undergoes no perturbation while capturing the image sequences, (2) focus of expansion (FOE) is theoretically constrained to be at the center of the image or close proximity for camera imperfection, and (3) the flow is radial and as a result the correspondence problem, so difficult elsewhere, reduces to a 1-D search problem.

For vision-based detection of lane boundaries, current computer vision algorithms for lane detection do not work well in regions of deep shadow. A pre-processing algorithm for detection and correction of shadow and low-sun angle artifacts was developed and successfully used to estimate lane boundaries on more than 6000 real highway images.

We describe a monocular feature tracker (MFT), the first stage of a low cost stereoscopic vision system for use on an autonomous guided vehicle (AGV) in an indoor environment. The system does not require artificial markings or other beacons, but relies upon accurate knowledge of the AGV motion. Linear array cameras (LAC) are used to reduce the data and processing bandwidths. The limited information given by LAC require modelling of the expected features. We model an obstacle as a vertical line segment touching the floor, and can distinguish between these obstacles and most other clutter in an image sequence. Detection of these obstacles is sufficient information for local AGV navigation.

The development of an autonomous mobile robot is a central problem in artificial intelligence and robotics. A vision system can be used to recognize naturally occurring landmarks located in known positions. The problem considered here is that of finding the location and orientation of a mobile robot using a 3-D image taken by a CCD camera located on the robot. The naturally occurring landmarks that we use are the corners of the room extracted by an edge detection algorithm from a 2-D image of the indoor scene. Then, the location and orientation of the vehicle are calculated by perspective information of the landmarks in the scene of the room where the robot moves.

In this paper a neural network approach to a particular class of scheduling problems is described. It uses a `k-cluster' network to enforce a non-preemptive time duration constraint and a `subset-sum' network to generate feasible schedules with a balanced load distribution. The two networks are super-imposed and the interactive nature of the parallel distributed processing inherent to Hopfield-Style neural networks negotiates the multiple constraint satisfaction requirements.

As the efficiency of road segmentation has a direct effect on the reliability of road following and planning -- and consequently the speed of the Autonomous Land Vehicle (ALV) -- road segmentation is one of the most preliminary and important tasks for the road following and planning of ALV, and a variety of methods for color road segmentation have been proposed. This presentation proposes a new data-fusion-based color road segmentation method in which a pyramid-based data structure and the corresponding region splitting and combination techniques for the classification of sensed areas are adopted. In the segmentation process, the roads are first segmented in two 1-D color spaces, and the data fusion technique is then used to combine the two classification results, improving the accuracy of the road segmentation.

The ability of a robot to find a goal in an unknown environment is significantly improved from previous path planners with the synthesis of intelligence and reflexive algorithms as a path planning approach. A hybrid expert system vector field (HESV) path planner is introduced and tested in 100 map scenarios using a custom robotic simulation shell. The method combines a trap detection expert system, a low-level reflexive obstacle avoidance algorithm, and a trap evasion expert system, to achieve improved performance without sacrificing computational efficiency. Simulation results and evaluations are presented for the potential field, vector field, and HESV techniques. HESV's expert systems have been simplified to illustrate the performance improvements that result from a small amount of intelligence. Although recommendations for the structure of the path planning system are given, the purpose of the paper is to show the effectiveness of intelligent rules in path planning, rather than to define the actual rules to be used.

The telemeter is a high precision positional tracking device that works by triangulation over the angular measurements provided by two or more telegoniometers. Each of the latter is composed of a TV camera with a teleobjective, looking at the working space through a pair of orthogonal mirrors. A problem with this kind of structure is the automation of the initial pointing of the telegoniometers on the target. We propose to add a third camera with a wide- angle lens framing the whole workspace to provide the line of sight on which the target must lie. In this work we present this `three-eyes' structure and discuss its calibration, as well as the problems related to the strategy for scanning the line of sight and for detecting the tracking condition. Two possible applications to robot qualification and autonomous navigation also are presented.

This paper presents formalisms for describing societies of cooperating behavior-based mobile robots, including the coordination between members of homogeneous teams, members of heterogeneous castes, assemblages of behaviors on individual robots, as well as perceptual strategies within primitive sensorimotor behaviors. This formal language is intended to facilitate proving properties about systems described in it.

Correlation of images has been used less in vision-based navigation. In this paper we present a novel method to estimate the orientation of the vehicle relative to the roadway by using image sequence. A parallel correlation algorithm is used to detect the difference between the current view of the roadway with its next view, and the orientation of the vehicle can be estimated reliably in real time. In order to account for figure variation caused by 3-D dynamic environment and perspective effect exerted by the camera system, a weighted correlation method has been developed based on planar motion assumption and reprojection transformation. This method has been implemented on PIPE, a pipelined image processing system, and tested in our campus roadway in combination with optical flow method for the detecting of moving objects. The advantage of the method is that the motion parameters can be extracted reliably without prerequisite for image sequence and no special road model is needed, as it adapts itself to rather complicated situations with other objects sharing the same environment.

One of the Department of Energy's (DOE) ongoing tasks is the storage and inspection of a large number of waste barrels containing a variety of hazardous substances. Martin Marietta is currently contracted to develop a robotic system -- the Intelligent Mobile Sensor System (IMSS) -- for the automatic monitoring and inspection of these barrels. The IMSS is a mobile robot with multiple sensors: video cameras, illuminators, laser ranging and barcode reader. We assisted Martin Marietta in this task, specifically in the development of image processing algorithms that recognize and classify the barrel labels. Our subsystem uses video images to detect and locate the barcode, so that the barcode reader can be pointed at the barcode.

This paper describes two six-legged robots with hexagonal architecture: the first one, of electropneumatical type, is controlled by a neurological information system based on signals captured by optical and ultrasonic sensors; the second one, of electrical type, is controlled on the basis of a detailed study (starting from the Song & Waldron's theorem) of various kinds of kinematical gaits and of the transitions from one gait to the other.

This paper introduces the Grumman Autonomous Long Life Vehicle (LLV) research platform currently under development within Grumman's Corporate Research Center. Topics to be discussed include the system architecture, mechanical design, the actuators, the sensor subsystems, the computing and control hardware, the software system, and safety issues. Current progress and plans for the project also are described.

This paper introduces a unified approach to trajectory planning and tracking for an industrial mobile robot subject to non-holonomic constraints. We show (1) how a smooth trajectory is generated that takes into account the constraints from the dynamic environment and the robot kinematics; and (2) how a general predictive controller works to provide optimal tracking capability for nonlinear systems. The tracking performance of the proposed guidance system is analyzed by simulation.

A line laser has been used to help navigation and obstacle detection for the indoor mobile robot PARIDE (Pavia autonomous robot for industrial environment). PARIDE is an A.M.R. based on a TRC platform that is being implemented using a subsumption architecture simulated through software on a 68030 board. The line is projected on the robot path and continually monitored using a CCD camera to detect line interruption or deviation. Laser, CCD camera, and frame grabber have been chosen in order to be commercially available and within the power requirements and physical dimensions of the mobile platform. The robot has to act in a human environment thus the laser must be safe (class 3A), visible (685 nm) and powerful to allow good quality images acquisition. Images are filtered, during acquisition, in order to simplify the low-level image processing.

For a broad variety of robotic applications, the critical issues are: proximity to workstation, tethered or radio communications, nature of environment (structured or unstructured), economics, maturity of technology, and teleoperated or autonomous control. The key issue is how to substitute the computer for the human operator. In the long term, different implementations of supervisory control will help simplify this transition. The effectiveness of supervisory control depends on the ability to add incremental autonomy.

In this paper we discuss our experience with porting our agent-based SmartyCat software from one computing environment, the Apple Macintosh, to another, the Silicon Graphics (SGI) Indigo. A significant part of the effort goes into converting the graphical user interface (GUI) for SAL (the SmartyCat Agent Language) from the proprietary Macintosh windowing system to the more open X-windows based system of the SGI. We also discuss our current effort in integrating a VME 68030 processor card to the original Z80-based on-board computing system. The evolutionary path taken by us in developing the SmartyCat architecture is described followed by a discussion of what was learned from our experience.

This work presents a system for the automatic selective harvesting of asparagus in open field being developed in the framework of the Italian National Project on Robotics. It is composed of a mobile robot, equipped with a suitable manipulator, and driven by a stereo-vision module. In this paper we discuss in detail the problems related to the vision module.

Domestic robots are promising examples of the application of robotics to personal life. There have been many approaches in this field, but no successful results exist. The problem is that domestic environments are more difficult for robots than other environments, such as factory floors or office floors. Consequently, conventional approaches using a model of human intelligence to design robots have not been successful. In this paper, we report on a prototyped domestic vacuum-cleaning robot that is designed to be able to handle complex environments. The control software is composed of two layers, both of which are generally inspired by behaviors of living creatures. The first layer corresponds to a dynamically reconfigurable system of behaviors implemented in the subsumption architecture. The ability of the robot to support alternate configurations of its behaviors provides the robot with increased robustness. We have conveniently labeled particular configurations as specific `emotions' according to the interpretation of observers of the robot's behavior. The second layer simulates the hormone system. The hormone system is modeled using state variables, increased or decreased by stimuli from the environment. The hormone condition selects the robot's most suitable emotion, according to the changing environments. The robot hardware is built of off-the-shelf parts, such as an embedded CPU, inexpensive home-appliance sensors, and small motors. These parts keep the total building cost to a minimum. The robot also has a vacuum cleaning function to demonstrate its capability to perform useful tasks. We tested the robot in our laboratory, and successfully videotaped its robust behaviors. We also confirmed the hormone system to enhance the robot's plasticity and lifelike quality.

A mobile robot for a ham industry has been developed. The features of the factory, limited environment, sliding floor, and the necessity for a high storage flexibility makes impossible the usage of conventional transport systems, conveyors, automatic guided vehicles, etc. The developed system permits us to integrate the pervious transport system, that was based on fork lift trucks, with the advantage that in the case of contingency an operator can drive the mobile robot as a fork lift truck. In this way, the transport and storage can be done using fork lift trucks, mobile robots, or both, and all of them controlled by a planning program running in a Sun Sparcstation10.

This paper provides a simulation-based understanding of the effects of three physical system properties upon ground vehicle tracking performance: (1) command quantization, (2) command data latency, and (3) actuation bandwidth. The proposed metric for evaluating effects of these properties was integrated absolute lateral error (IAE). Under this metric, system tracking performance was optimized by (1) minimizing the data latency, (2) maximizing the actuation bandwidth, and (3) minimizing the command quantization, in that order. This work drew upon existing work (using a simple simulation model) with the intent of providing guidance for future efforts.

Outdoor mobile robot path tracking for an extended period of time and distance is a formidable task. The difficulty lies in the ability of robot navigation systems to reliably and accurately report on the position and orientation of the vehicle. This paper addresses the accurate navigation of mobile robots in the context of non-line of sight autonomous convoying. Dead-reckoning, GPS and vision based autonomous road following navigational schemes are integrated through a Kalman filter formulation to derive mobile robot position and orientation. The accuracy of these navigational schemes and their sufficiency to achieve autonomous path tracking for long duration are examined.

This paper discusses some communications issues relevant to a system consisting of an arbitrarily large number of simple autonomous robots, from several different perspectives: (a) in the language of traditional communications systems analysis, (b) by analogy with biological systems, and (c) as a problem of intelligent control and data fusion. Emphasis is placed on support for the system development process: providing the developer with the ability to efficiently download software revisions and to determine the actual internal behavioral states of system elements while in operation. Also treated are the need to support coordination (or at least cooperation) between elements during system operation, and the process by which an external user can input commands to the system and to receive back from the system such information as system status, environmental characterization, and indications of mission progress. An approach to implementation of a low-cost testbed environment is also presented.

A goal of the Surrogate Semi-Autonomous Vehicle (SSV) program is to have multiple vehicles navigate autonomously and cooperatively with other vehicles. In this paper we address the steps to develop the global navigation system (GNS) for the SSV. We also discuss GNS components, specifications, and requirements for meeting SSV system needs. The development process, results, lessons learned, and remaining issues are discussed. We selected a low-cost solution because no available integrated global positioning system/inertial navigation system (GPS/INS) existed at the time of selection.

In two previous papers we explored some of the systems aspects of applying large numbers of inexpensive robots to real world applications. The concept of coverage can help the user of such a system visualize its overall function and performance in mission-relevant terms, and thereby support necessary system command control functions. An important class of coverage applications are those that involve a search, in which a number of searching elements move about within a prescribed search area in order to find one or more target objects, which may be stationary or mobile. A simple analytical framework was employed in the previous work to demonstrate that the design of a cost-effective many-robot search system can depend sensitively on the interplay of sensor cost and performance levels with mission-specific functional and performance requirements. In the current paper we extend these results: we consider additional measures of effectiveness for area search systems to provide a broader basis for a tradeoff of coordinated versus random search models, and we explore how to deliberately achieve effectively randomized search strategies that provide uniform search coverage over a specified area.

This paper introduces a proposed approach for robust exterior autonomous navigation, based on the use of redundant, and complimentary position estimation sensors, fused and integrated to provide robust operation. The proposed approach considers constraints due to a combination of theoretical and practical limitations of the sensors performance characteristics, and sensor fusion and integration algorithm limitations. A hypothetical exterior autonomous navigation application is also introduced to illustrate a systems perspective on one class of autonomous navigation, specifically routine autonomous navigation in a known, semi-structured environment. The focus of this paper is primarily on the autonomous vehicles position estimation sensor suite selection, and sensor fusion and integration approach.

The Naval Command Control and Ocean Surveillance Center (NCCOSC) has developed an architecture to provide coordinated control of multiple autonomous vehicles from a single host console. The Multiple Robot Host Architecture (MRHA) is a distributed, LAN-based, multiprocessing system that can be expanded to accommodate as many as 32 robots. The initial application will employ eight Cybermotion K2A Navmaster robots configured as remote security platforms in support of the Mobile Detection Assessment and Response System (MDARS) Program. MDARS is a joint Army-Navy development effort which seeks to provide an automated intrusion detection and inventory assessment capability for use in DoD warehouses and storage sites.

Although numerous commercial opportunities exist for automated mobility solutions, few have been realized. Mobile robot development has faltered primarily due to the absence of high quality navigation data. This paper presents CONAC (computerized opto-electronic navigation and control), a solution to the navigation gap that offers a comprehensive approach to automated mobility. The major breakthrough is the development of a practical, real-time navigation technology that is thousands of times more accurate than GPS and is usable indoors or outdoors.