ELVIS: Eigenvectors for Land Vehicle Image System

Hancock, John; Thorpe, C.

doi:10.1109/iros.1995.525772

Cited by 18 publications

(15 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, we implemented eigenvector projection using principal component analysis developed for road-following using video imagery [17]. It is similar to steering using a neural network [16], but assumes a linear model of the control equation.…”

Section: Resultsmentioning

confidence: 99%

“…Typically "learning" applies to perception as in learning a map of the environment [12] or learning a classification of the terrain given image or range data [13]- [15]. In a few cases learning is applied to the control itself as in a vehicle that learns to follow roads based on onboard video cameras while a human drives [16], [17]. We are not aware of other work where the robot learns to avoid obstacles based on observation of a human driver.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Learning to Drive Among Obstacles

Hamner

Scherer

Singh

2006

2006 IEEE/RSJ International Conference on Intelligent Robots and Systems

View full text Add to dashboard Cite

Abstract-This paper reports on an outdoor mobile robot that learns to avoid collisions by observing a human driver operate a vehicle equipped with sensors that continuously produce a map of the local environment. We have implemented steering control that models human behavior in trying to avoid obstacles while trying to follow a desired path. Here we present the formulation for this control system and its independent parameters, and then show how these parameters can be automatically estimated by observation of a human driver. We present results from experiments with a vehicle (both real and simulated) that avoids obstacles while following a prescribed path at speeds up to 4 m/sec. We compare the proposed method with another method based on Principal Component Analysis, a commonly used learning technique. We find that the proposed method generalizes well and is capable of learning from a small number of examples.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Learning to Drive Among Obstacles

Hamner

Scherer

Singh

2006

2006 IEEE/RSJ International Conference on Intelligent Robots and Systems

View full text Add to dashboard Cite

show abstract

“…Sense data includes resolved force/torque components and the position information of the robot itself. For our experiments there were no redundant groups of very similar and correlated sense elements such as a visual retina (as in [Hancock and Thorpe, 1995]) or sonar ring (as in [Pierce, 1991]) as has been employed in other eigenspace approaches. The only structure imposed at all on the learning process is linearity, so the result learned…”

Section: 2adapting the Expertise Database Through Learningmentioning

confidence: 99%

Gesture-based programming for robotics: human-augmented software adaptation

Voyles¹,

Morrow²,

Khosla³

1999

IEEE Intell. Syst.

View full text Add to dashboard Cite

Gesture-Based Programming is a paradigm for programming robots by human demonstration in which the human demonstrator directs the self-adaptation of executable software. The goal is to provide a more natural environment for the user as programmer and to generate more complete and successful programs by focusing on task experts rather than programming experts. We call the paradigm "gesture-based" because we try to enable the system to capture, in real-time, the intention behind the demonstrator's fleeting, context-dependent hand motions, contact conditions, finger poses, and even cryptic utterances in order to reconfigure itself. The system is self-adaptive in the sense that knowledge of previously acquired skills (sensorimotor expertise) is retained by the system and this knowledge facilitates the interpretation of the gestures during training and then provides feedback control during run-time. INTRODUCTIONThe text-based programming paradigm relies exclusively on the expertise of the human programmer and his or her ability to learn from and remember past lessons on system development. Beyond on-line help files, the applications themselves and the programming environments that facilitate their creation acquire no knowledge of themselves and carry-over no useful knowledge from project to project or even within the development of a single project. The burden of retaining the expertise that results from the creation of new applications or new configurations of existing applications falls entirely on the human programmer. The idea behind self-adaptive software is to empower the system itself to participate in its own development. If a software environment can retain a knowledge base of its own capabilities and usefulness, it can assist the human developer during reapplication or reconfiguration and potentially perform autonomous adaptation. The allowable degree of "self-determination" will certainly vary across applications and operating environments. For example, the user interface to a consumer application program may have wide latitude for self-adaptation on an ongoing basis. A manufacturing system, on the other hand, would likely have rigid constraints on the degree of allowable self-adaptation and even on the periods during which self-adaptation is enabled. GESTURE-BASED PROGRAMMING FOR ROBOTICSThe focus of this paper is not the grand goal of self-adaptive software, but a smaller step toward it we call human-augmented adaptation. We further limit our efforts to the task of robot programming. In particular, we are interested in the programming paradigm itself and in finding alternative approaches to programming that are more intuitive for human users. Our basic model is human-to-human training, which is very intuitive to users. In this model, the trainee is active in the process, so, when applied to the programming of robotic systems, it naturally leads to systems that are more participative in their own development. In this sense, human-augmented adaptation is an important step toward, and a useful compl...

show abstract

“…The shape from motion technique is modified to match the principal component analyses of Hancock and Thorpe [4] and Pierce [7]. The standard principal components analysis provides a substitute for the geometric constraint used in equation (14).…”

Section: Disparate Sensory Datamentioning

confidence: 99%

“…The training data was batch-processed using SVD to extract the eigenvectors and selected the largest n eigenvectors using the largest ratio of adjacent singular values as the threshold for n. For run-time operation (sensor fusion in real-time), the new sensor image is projected onto the eigenspace as described in [4].…”

Section: Disparate Sensory Datamentioning

confidence: 99%

Including sensor bias in shape from motion calibration and sensor fusion

Voyles

Merrow²,

Khosla³

1996 IEEE/SICE/RSJ International Conference on Multisensor Fusion and Integration for Intelligent Systems (Cat. No.96TH8242)

View full text Add to dashboard Cite

Shape from Motion data fusion brings a greater degree of autonomy and sensor integration to intelligent systems in IntroductionThe Shape from Motion sensor fusion technique originated from a basic need: more efficient sensor calibration strategies. We initially wanted a quicker method to calibrate a multitude of sensors for multi-fingered robotic and human hands. After developing this novel technique for sensor calibration that eliminates much of the need for known applied loads, we realized it could be used in a more general framework for data fusion involving constant linear transformations. While it is possible to reformulate the Shape from Motion approach for incremental rather than batch processing, it will not be discussed here.In an abstract sense, calibration involves extracting (or "learning") the "shape" of the linear mapping between the input and output spaces of groups of sensors. The first sections of this paper describe Shape from Motion calibration and the extensions to sensor bias in this context. Section 9 describes a "primordial learning" application in which a mobile robot uses disparate data fusion to learn to free itself from jams. Motivation for Shape from MotionMulti-axis force/torque sensor calibration --the process of recovering the linear calibration matrix mapping raw sensor values to resolved forces and torques --is a data fusion problem that has traditionally been a very costly procedure in terms of time. The reason for this is the traditional methods based on reduction by least squares require two sets of data: input and output. Output data is very easy to gather so incurs little cost. Input data, on the other hand, can be quite expensive.The output data consists of a vector of raw sensor values. For common wrist force/torque sensors based on a maltese cross configuration, such as the JR3 or Lord commercial products, this vector consists of eight strain gauge values. Collecting this vector is a simple matter of data acquisition.The input data set is quite different. It represents the applied forces and torques that produce each and every output vector. For multi-axis sensors, this is a complex combination of vectors that must be carefully measured with respect to an external reference frame. These careful measurements in 3-space are very time consuming but are of utmost importance in recovering an accurate calibration matrix.In an ideal world, n separate measurements are required for an n-dimensional sensor. However, there are two primary sources of noise: measurement noise in acquiring the output and inaccuracy in applying the input. In order to recover an accurate calibration matrix, it is desirable to gather as much data as possible; all things being equal, more data results in greater accuracy. Unfortunately, each piece of data comes at high cost due to the input.Shape from Motion calibration is a new technique that virtually eliminates the need for the input data set. Instead of gathering both inputs and outputs, outputs only are gathered. Because collecting outputs is...

show abstract

ELVIS: Eigenvectors for Land Vehicle Image System

Cited by 18 publications

References 5 publications

Learning to Drive Among Obstacles

Learning to Drive Among Obstacles

Gesture-based programming for robotics: human-augmented software adaptation

Including sensor bias in shape from motion calibration and sensor fusion

Contact Info

Product

Resources

About