A key problem in robot grasping is that of positioning the manipulator contacts so that an object can be grasped. In unstructured environments, contact positions are typically planned based on range or visual measurements that are used to reconstruct object geometry. However, because it is difficult to measure the complete object geometry precisely in common grasp scenarios, it is useful to employ additional techniques to adjust or refine the grasp using local information only. In particular, grasp control techniques can be used to improve a grasp by adjusting the contact configuration after making initial contact with an object by using measurements of local object geometry at the contacts. This paper proposes three variations on null-space grasp control: an approach that combines multiple grasp objectives to improve a grasp. Two of these variations are theoretically demonstrated to converge to force-closure configurations for arbitrary convex objects when grasping with two contacts. All variations are found to converge in simulation. Robot-grasping experiments are reported that show the approach to be useful in practice.

When grasping an object, a robot must identify the available forms of interaction with that object. Each of these forms of interaction, a grasp affordance, describes one canonical option for placing the hand and fingers with respect to the object as an agent prepares to grasp it. The affordance does not represent a single hand posture, but an entire manifold within a space that describes hand position/orientation and finger configuration. Our challenges are 1) how to represent this manifold in as compact a manner as possible, and 2) how to extract these affordance representations given a set of example grasps as demonstrated by a human teacher. In this paper, we approach the problem of representation by capturing all instances of a canonical grasp using a joint probability density function (PDF) in a hand posture space. The PDF captures in an object-centered coordinate frame a combination of hand orientation, grasp centroid position and offset from hand to centroid. The set of canonical grasps is then represented using a mixture distribution model. We address the problem of learning the model parameters from a set of example grasps using a clustering approach based on expectation maximization. Our experiments show that the learned canonical grasps correspond to the functionally different ways that the object may be grasped. In addition, by including the grasp centroid/hand relationship within the learned model, we eliminate this as a hard-coded parameter and the resulting approach is capable of separating different grasp types, even when the different types involve similar hand postures.

We describe a method for predicting user intentions as part of a human-robot interface. In particular, we show that funnels, i.e., geometric objects that partition an input space, provide a convenient means for discriminating individual objects and for clustering sets of objects for hierarchical tasks. One advantage of the proposed implementation is that a simple parametric model can be used to specify the shape of a funnel, and a straightforward heuristic for setting initial parameter values appears promising. We discuss the possibility of adapting the user interface with machine learning techniques, and we illustrate the approach with a humanoid robot performing a variation of a standard peg-insertion task.

In dexterous manipulation, an object must be reconfigured while maintaining a stable grasp. This may require that the object be re-grasped in order to avoid finger workspace limits. We present a set of closed-loop controllers designed to achieve force-related objectives such as wrench closure, and show how they may be concurrently combined. Furthermore, we show that dexterous manipulation behavior may be generated by sequencing concurrent combinations of these controllers. We show that dexterous manipulation can be viewed as a task that is accomplished in the context of a wrench closure constraint. We hypothesize this approach can generalize to any task that must be accomplished while maintaining a set of constraints.

Learning techniques in robotic grasping applications have usually been concerned with the way a hand approaches to an object, or with improving the motor control of manipulation actions. We present an active learning approach devised to face the problem of visually-guided grasp selection. We want to choose the best hand configuration for grasping a particular object using only visual information. Experimental data from real grasping actions is used, and the experience gathering process is driven by an on-line estimation of the reliability assessment capabilities of the system. The goal is to improve the selection skills of the grasping system, minimizing at the same time the cost and duration of the learning process.

In large scale surveillance applications, coherent presentation of data coming from myriad sensors becomes a problem. For example, tasks such as "locate an intruder" are no longer easy when the user is facing a room of monitors connected to hundreds of cameras. Therefore, there is a need for a system that allows the user to easily navigate the data space. Due to the scale of the application, such systems should also be robust with respect to hardware and software failures, as well as to varying bandwidth conditions. Our strategy is to first build software units that provide sensor abstractions (e.g., location of a person, noise level of a predefined region) to lessen the burden of dealing with individual sensors from the user. Second, we project this abstract sensor information into an augmented virtual reality interface for presentation to the user. The AVR interface offers a common medium to display abstract information constructed from multiple sensors, as well as allowing access to raw sensor information such as video streams, or a mixture of both. Also, the AVR interface can synthesize views not serviced by the physical cameras. The sensor abstraction and the smooth transition between sensors enable the user to intuitively navigate the data space. Further more, through this interface, the user can dynamically reconfigure the system resources. We will demonstrate three scenarios highlighting the above mentioned features.

Although it is mechanically possible for a robot manipulator to grasp using non-fingertip contacts, there are few examples of this. We call non-fingertip grasping such as grasping with proximal finger phalanges or grasping with the sides of arms "whole body grasping". While robotic demonstrations are rare, humans commonly use whole body grasps to interact with the world. One of the distinctive features of whole body grasping is the kinematic coupling among potential contacts. This kinematic coupling introduces extra constraints into the grasp synthesis problem. In this paper, we extend recent grasp control techniques to whole body grasping. We show how the grasp control may be parameterized with a set of contact points on the surface of the robot manipulator that enables the grasp search to handle the extra kinematic coupling constraints and find whole body grasps.

This paper deals with visually guided grasping of unmodeled objects for robots which exhibit an adaptive behavior based on their previous experiences. Nine features are proposed to characterize three-finger grasps. They are computed from the object image and the kinematics of the hand. Real experiments on a humanoid robot with a Barrett hand are carried out to provide experimental data. This data is employed by a classification strategy, based on the k-nearest neighbour estimation rule, to predict the reliability of a grasp configuration in terms of five different performance classes. Prediction results suggest the methodology is adequate.

With the deployment of large, distributed networks of cameras and other sensors, it is becoming necessary to also address the issue of how to effectively present the large volume of gathered information to a user. The paper presents a virtual/augmented reality architecture that has been explicitly designed for use with a fully-portable, wearable computing system. A critical component of this system is a network-based mechanism for the representation of virtual objects and the live communication of changes in their state to users located elsewhere on the network. By presenting virtual objects in a uniform manner over the network, it becomes easy to construct new dynamic, virtual environments that reflect the state of robots or humans within the real environment. We demonstrate the utility of the architecture through several robot and human tracking examples.

Research in grasping and manipulation has mainly focused on the theoretical framework of the grasp analysis and synthesis, however there are few works in the practical implementation of these approaches. We propose several methods and criteria for constraining visually-selected planar grasps to the particular kinematics of a three finger gripper, the Barrett Hand. The method relies on the features of an object provided by a vision system, and uses several criteria for classifying the possible grasps according to the configurations of the gripper that fit them. It has been implemented on the UMass Torso and the results are discussed.

Much of the tradition in robot grasping is rooted in geometrical, planning-based approaches in which it is assumed that object geometries are well modeled a priori. Some recent approaches have chosen instead to deal with objects of unknown geometry. These techniques treat grasping as an active sensory-driven problem. At any given time, finger contacts are incrementally displaced along the object's local surface using a single control law. In this paper, we extend this approach by allowing multiple control laws to be active simultaneously. Three control laws are combined by projecting the actions of subordinate control laws into other control law nullspaces. The resulting composite controller finds grasps that are more robust than the component primitives in isolation. Finally, we show how this approach may be used on hand/arm manipulation systems with arbitrary kinematics.

A set of impedance controllers is used for both state estimation and tracking control on a mobile robot. State estimation is based on the states of a family of impedance controllers and tracking is implemented through a single controller from this set. Reinforcement learning techniques are used to create switching policies that optimize time or energy in a path tracking task.

A simplified model of the cerebellum was developed to explore its potential for adaptive, predictive control based on delayed feedback information. An abstract representation of a single Purkinje cell with multistable properties was interfaced, using a formalized premotor network, with a simulated single degree-of-freedom limb. The limb actuator was a nonlinear spring-mass system based on the nonlinear velocity dependence of the stretch reflex. By including realistic mossy fiber signals, as well as realistic conduction delays in afferent and efferent pathways, the model allowed the investigation of timing and predictive processes relevant to cerebellar involvement in the control of movement. The model regulates movement by learning to react in an anticipatory fashion to sensory feedback. Learning depends on training information generated from corrective movements and uses a temporally asymmetric form of plasticity for the parallel fiber synapses on Purkinje cells.

Biological control systems have long been studied as possible inspiration for the construction of robotic controllers. The cerebellum is known to be involved in the production and learning of smooth, coordinated movements. In this paper, we present a model of cerebellar control of a muscle-actuated, two-link, planar arm. The model learns in a trial-and-error fashion to produce bursts of muscle activity that accurately bring the arm to a specified target. When the cerebellum fails to bring the arm to the target, an extra-cerebellar module performs four-quality corrective movements, from which the cerebellum may update its program. In learning to perform the task, the cerebellum constructs an implicit inverse model of the plant. This model uses a combination of delayed sensory signals and recently-generated motor commands to compute the new output motor signal.

Within the field of robotics, much recent attention has been given to control techniques that have been termed reactive or behavior-based. The design of such control systems for even a remotely interesting task is typically a laborious effort, requiring many hours of experimental "tweaking" as the actual behavior of the system is observed by the system designer. In this paper, the authors present a neural-based reinforcement learning approach to the design of reactive control policies in which the designer specifies the the desired behavior of the system, rather than the control program that produces the desired behavior.>

A control system architecture is described for an autonomous flying vehicle. The vehicle, equipped with fourteen sensors, uses a model helicopter as an airframe. The control system utilizes these sensors to: remain aloft and in stable flight; navigate to a target; and manipulate a physical object. The overall approach to the problem is based on a behavioral paradigm. The key contribution of this paper is the demonstration of a situated agent under these severe circumstances; as the craft is airborne, it is in constant risk of crashing. Unlike terrestrial mobile robots, the craft must constantly make sound decisions to maintain its integrity.

The authors describe the staged evolution of a complex motor pattern generator (MPG) for the control of a walking robot. The experiments were carried out on a six-legged, Brooks-style insect robot. The MPG was composed of a network of neurons with weights determined by genetic algorithm optimization. Staged evolution was used to improve the convergence rate of the algorithm. First, an oscillator for the individual leg movements was evolved. Then, a network of these oscillators was evolved to coordinate the movements of the different legs. By introducing a staged set of manageable challenges, the algorithm's performance was improved.>

R/sup 2/AD, a flexible robotics application development environment, combines powerful graphical display features with tools that support the rapid development of application programs. The graphical interface allows researchers to monitor the system state variables during the dynamic execution of robot programs. This feature enhances the researcher's efficiency in interpreting experimental results. Rapid development is supported by a virtual machine through the use of software engineering principles and object-oriented techniques. In addition to these features, the user is able to emulate a wide variety of robot computational paradigms. This allows experimentation with, and comparison of, various computational techniques (e.g., hierarchical versus heterarchical control process topologies). The utility of the R/sup 2/AD virtual machine is demonstrated through applications currently under development.>