Autonomous Land Vehicle Research Articles

A Bootstrap Mode for An Autonomous Vehicle

A new modular hardware design for an autonomous land vehicle is described, together with a “bootstrap” algorithm, which will initialise the vehicle with respect to its driving corridor. The bootstrap method detects the road's vanishing point by matching plan-view transformed horizontal bands of the road image, and uses the result to realign the vehicle with the road edge if required. Edge detection is used to find parallel lines, in the real world, that pass through the vanishing point and white line detection determine boundaries and the correct road edges

Knowledge-Based Vehicle Navigation in Complex Road Networks

This paper presents a computer vision system designed for autonomous land vehicle navigation. The goal of the system is to enable the vehicle to drive automatically along a given complex road network whose geometrical description is known. Roads are detected from a sequence of monochromatic images captured by a video camera. A knowledge-based approach is adopted. High-level knowledgeabout road networks, as well as the speed and location of the vehicle, is used as the mam source of informationn to guide the vehicle

Dynamic-scene and motion analysis using passive sensors I. A qualitative approach

Some dynamic-scene and motion analysis techniques developed to support the DARPA Strategic Computing program's autonomous land vehicle effort are described. The approach, called dynamic reasoning from integrated visual evidence (DRIVE), addresses the key problems of estimating robot motion from visual cues, detecting and tracking moving objects, and constructing and maintaining a global dynamic reference model. With reliable computation of displacement fields, this qualitative technique for understanding dynamic scenes reasons accurately on hundreds of image frames. The technique can also be extended to cases where the features are lines, regions, or contours. >

Dynamic-scene and motion analysis using passive sensors. II. Displacement-field and feature-based approaches

For pt.I see ibid., vol.7, no.1, p.45-52 (1992). Dynamic-scene and motion analysis techniques developed to support the DARPA Strategic Computing program's autonomous land vehicle effort are described. Motion-understanding techniques using displacement fields, 3-D motion and structures, and feature correspondence, which show promise, are discussed, and the issues to be addressed and problems to be solved before robust dynamic-scene and motion analysis is achieved are examined.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Position estimation for an autonomous mobile robot in an outdoor environment

The position estimation problem for an autonomous land vehicle navigating in an unstructured mountainous terrain is solved. A digital elevation map (DEM) of the area is assumed to be given. It is also assumed that the robot is equipped with a camera that can be panned and tilted, a compass, and an altimeter. No recognizable landmarks are assumed to be present, and the robot is not assumed to have an initial estimate of its position. The solution structures the problem as a constrained search paradigm by searching the DEM for the possible robot location. The shape and position of the horizon line in the image plane and the known camera geometry of the perspective projection are used as parameters. Geometric constraints are used to prune the search space significantly. The algorithm is made robust to errors in the imaging process by accounting for worst case errors. The approach is tested using real terrain data of areas in Colorado and Texas.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Collision avoidance by a modified least‐mean‐square‐error classification scheme for indoor autonomous land vehicle navigation

AbstractIn this article, a new collision‐avoidance scheme is proposed for autonomous land vehicle (ALV) navigation in indoor corridors. The goal is to conduct indoor collisionfree navigation of a three‐wheel ALV among static obstacles with no a priori position information as well as moving obstacles with unknown trajectories. Based on the predicted positions of obstacles, a local collision‐free path is computed by the use of a modified version of the least‐mean‐square‐error (LMSE) classifier in pattern recognition. Wall and obstacle boundaries are sampled as a set of 2D coordinates, which are then viewed as feature points. Different weights are assigned to different feature points according to the distances of the feature points to the ALV location to reflect the locality of path planning. The trajectory of each obstacle is predicted by a real‐time LMSE estimation method. And the maneuvering board technique used for nautical navigation is employed to determine the speed of the ALV for each navigation cycle. Smooth collision‐free paths found in the simulation results are presented to show the feasibility of the proposed approach.

Autonomous cross-country navigation: an integrated perception and planning system

The development of a software system for obstacle avoidance that controls a vehicle smoothly, continuously, and safely across natural terrain en route to a goal is discussed. The work, carried out as part of the Strategic Computing Initiative of the US Defense Advanced Research Projects Agency (DARPA), was performed on the Autonomous Land Vehicle at Martin Marietta's Denver facilities. The planning software uses digital maps to plan a preferred route, and then, as the vehicle traverses that route, sensors are used to obtain scene descriptions. Techniques necessary to detect and avoid obstacles in cross-country terrain, with minimum delay between sensing and acting, are used. The system architecture, the cross-country perception system, the planning system for obstacle avoidance, the simulation environment, and the experiments are described. Work to develop a fine-grained control architecture, internalized plans, multisensor perception, and color image segmentation is discussed. >

The Warp machine on Navlab

The authors review the history of the Carnegie-Mellon Warp machine on Navlab, an autonomous land vehicle, and describe three Navlab vision systems implemented on the Warp machine. They then critically evaluate components of Warp in light of this experience. The Warp machine was used to implement stereo vision for obstacle avoidance and color-based road-following systems. The stereo-vision system was FIDO, which is derived from some of the earliest work in vision-guided robot vehicle navigation. Two color-based road following systems were implemented; one adapted conventional vision techniques to the problem of road recognition and the other used a neural network-based technique to learn road following online. Finally, the authors discuss the value of applications integration with machine development, discuss the limitations of the attached processor model, and give recommendations for future systems.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Efficient Training of Artificial Neural Networks for Autonomous Navigation.

The ALVINN (Autonomous Land Vehicle In a Neural Network) project addresses the problem of training artificial neural networks in real time to perform difficult perception tasks. ALVINN is a backpropagation network designed to drive the CMU Navlab, a modified Chevy van. This paper describes the training techniques that allow ALVINN to learn in under 5 minutes to autonomously control the Navlab by watching the reactions of a human driver. Using these techniques, ALVINN has been trained to drive in a variety of circumstances including single-lane paved and unpaved roads, and multilane lined and unlined roads, at speeds of up to 20 miles per hour.

ALV Navigation Using Terrain Databases

A navigation strategy for an autonomous land vehicle (ALV) based on the Defence Geographic Database is currently being developed under the auspices of. the Mobile Advanced Robotics Defence Initiative (MARDI). The database is hierarchically structured, and supports the storage of an unlimited number of attributes for each geographical feature. It also expressly identifies the interconnectivity of adjacent features, making it ideal for automatic route calculation. Tactical battlefield information is being combined with the database resulting in a composite link-node map which will be searched for optimum routes using standard algorithms. Both military and geographical criteria will be simultaneously optimized. Waypoints suitable for machine vision techniques will also be extracted and used to confirm the ALV location.

Rigid body motion from range image sequences

An algorithm is described for recovering the six degrees of freedom of motion of a vehicle from a sequence of range images of a static environment taken by a range camera rigidly attached to the vehicle. The technique utilizes a least-squares minimization of the difference between the measured rate of change of elevation at a point and the rate predicted by the so-called elevation rate constraint equation. It is assumed that most of the surface is smooth enough so that local tangent planes can be constructed, and that the motion between frames is smaller than the size of most features in the range image. This method does not depend on the determination of correspondences between isolated high-level features in the range images. The algorithm has been successfully applied to data obtained from the range imager on the Autonomous Land Vehicle (ALV). Other sensors on the ALV provide an initial approximation to the motion between frames. It was found that the outputs of the vehicle sensors themselves are not suitable for accurate motion recovery because of errors in dead reckoning resulting from such problems as wheel slippage. The sensor measurements are used only to approximately register range data. The algorithm described here then recovers the difference between the true motion and that estimated from the sensor outputs.

Road shape reconstruction by local flatness approximation

A new algorithm is presented for reconstructing the three-dimensional (3-D) road shape from camera images for the purposes of navigating autonomous land vehicles. The approximation that the road surface is locally flat enables us to determine a one-to-one correspondence between the two road boundaries, which together with our knowledge about roads (the 'model' of roads) determines the 3-D road shape. In order to cope with the inaccuracy of image data, a least-squares curve-fitting technique is proposed, and the behaviours of image noise are analysed. Some examples based on real images are given.

Implementing task-level mission management for intelligent autonomous vehicles

Mission Management forms an integral part of the command and control system for truly intelligent autonomous vehicles. It is responsible for overall coordination of robot subsystem activities, directing them to achieve some optimal set of goals. The Mission Management system for an autonomous land vehicle is implemented in C ++ using clausal form logic and temporal reasoning. The representations used and the structure of the software are described, emphasising the separation in software of the planning representations from their implementation in C ++ classes

Computing camera parameters using vanishing-line information from a rectangular parallelepiped

A new approach to camera calibration using vanishing line information for three-dimensional computer vision is proposed. Calibrated parameters include the orientation, the position, the focal length, and the image plane center of a camera. A rectangular parallelepiped is employed as the calibration target to generate three principal vanishing points and then three vanishing lines from the projected image of the parallelepiped. Only a monocular image is required for solving these camera parameters. It is shown that the image plane center is the orthocenter of a triangle formed by the three vanishing lines. From the slopes of the vanishing lines the camera orientation parameters can be determined. The focal length can be computed by the area of the triangle. The camera position parameters can then be calibrated by using related geometric projective relationships. The derived results show the geometric meanings of these camera parameters. The calibration formulas are analytic and simple to compute. Experimental results show the feasibility of the proposed approach for a practical application—autonomous land vehicle guidance.

A flexible tool for prototyping ALV road following algorithms

A production system model of problem-solving is applied to the design of a vision system by which an autonomous land vehicle (ALV) navigates roads. The ALV vision task consists of hypothesizing objects in a scene model and verifying these hypotheses using the vehicle's sensors. Object hypothesis generation is based on the local navigation task, an a priori road map, and the contents of the scene model. Verification of an object hypothesis involves directing the sensors towards the expected location of the object, collecting evidence in support of the object, and reasoning about the evidence. Constructing the scene model consists of building a semantic network of object frames exhibiting component, spatial, and inheritance relationships. The control structure is provided by a set of communicating production systems implementing a structured blackboard; each production system contains rules for defining the attributes of a particular class of object frame. The combination of production system and object-oriented programming techniques results in a flexible control structure able to accommodate new object classes, reasoning strategies, vehicle sensors, and image analysis techniques. >

Efficient algorithms for obstacle detection using range data

Algorithms are presented which segment range images and classify regions as being navigable or unnavigable by a land vehicle. The partial derivatives in spherical coordinates of range images are shown to yield rapid and robust results for obstacle detection. The algorithms are applied to data collected from an active laser range sensor mounted on an autonomous land vehicle and their comparative results are analyzed. The sensitivity of various algorithms to uncertainty in the orientation of the range sensor are studied. Experiments on image enhancement, including iterative symmetric-nearest-neighbor smoothing and mixed-pixel minimization, are also presented.

A KNOWLEDGE-BASED NAVIGATION SCHEME FOR AUTONOMOUS LAND VEHICLES

A knowledge-based navigation system for autonomous land vehicles (ALVs) has been developed which can successfully negotiate an obstacle and threat-laden terrain, even if nothing is known beforehand about the terrain. The ALV stores new information in its memory as it travels, has the ability to backtrack out of unexpected dead ends, and performs spontaneous decision making in the field based on local sensor readings. The optimal global route of the ALV journey is obtained using dynamic programming, and decision making is accomplished via a production rule-based system. Execution examples demonstrate the power of the prototype system to solve navigation problems. This establishes the feasibility of constructing a valid ALV by combining search techniques with artificial intelligence tools such as production rule-based systems.

Range-video fusion and comparison of inverse perspective algorithms in static images

Creating three-dimensional (3-D) descriptions of road boundaries from single 2-D images of segmented road is a central problem for the autonomous land vehicle (ALV) road-following task. It is shown how heuristic simplifying assumptions can be avoided by using actively scanned laser range data to determine the 3-D space location of features in a 2-D image. An epipolar-plane approach that is commonly used to restrict search spaces in stereo correspondence problems is used to combine data gathered from the ALV's video camera and an ERIM laser range scanner for accurate 3-D descriptions of roads. In addition, various heuristic inverse perspective algorithms are compared with each other and with the video/range scanner fusion approach in terms of accuracy, failure situations, and estimated computational speed. The simplest of these is the flat-earth algorithm where the assumption is made that the vehicle is at all times resting in a level attitude on an infinite flat plane; image points were projected onto this plane. A second approach, the hill-and-dale algorithm, permits the plane's elevation to vary such that image pairs of left and right road edge points define a road of constant width. The zero-bank algorithm imposes a constant-road-width constraint on heuristically extracted image pairs of road edge points and the algorithm seeks image pairs of edge points that correspond to true 'crossword' road edges on curving roads. Graphical output describing the performance of the various algorithms on real-world scenes is presented and discussed.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

局所平面近似による道路形状復元

A new algorithm is presented for reconstructing the 3 D road shape from camera images for the purpose of navigating autonomous land vehicles (ALVs) . The apporoximation that the road surface is locally flat enables us to determine a one-to-one correspondence between the two road boundaries, which together with our knowledge about the road shape (the “model” of roads) determines the 3 D road shape. In order to cope with inaccuracy of image data, a least-squares curve fitting technique is proposed. An example based on real image data is also shown.

Model-directed mobile robot navigation

The authors report on the system and methods used by UMass Mobile Robot Project. Model-based processing of the visual sensory data is the primary mechanism used for controlling movement of an autonomous land vehicle through the environment, measuring progress towards a given goal, and avoiding obstacles. Goal-oriented navigation takes place through a partially modeled, unchanging environment that contains no unmodeled obstacles; this simplified environment provides a foundation for research in more complicated domains. The navigation system integrates perception, planning, and execution of actions. Of particular importance is that the planning processes are reactive and reason about landmarks that should be perceived at various stages of task execution. Correspondence between image features and expected landmark locations are used at several abstraction levels to ensure proper plan execution. The system and some experiments that demonstrate the performance of its components is described.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>