We report on lift experiments conducted at scale for an artificial platform mimicking the dragonfly species: Sympetrum sanguineum. The platform, as well as the lift sensor, was custom designed and built. The flapping mechanism consisted of a piezoelectric bending-beam actuator, transmission using carbon-fiber elements and polymide-film joints, and wings constructed of polyester film with carbon-fiber support structure. The flapping kinematics of the Sympetrum san-guineum was replicated as closely as possible although only a pair of forewings were used in these experiments. The lift generated, when accounting for the addition of a pair of hindwings, is sufficient to allow for the hovering of a real-life dragonfly. The results, the first at-scale fully transient measurements of artificial dragonfly forewings, show that the lift curves quantitatively as well as qualitatively validate existing 2D and 3D computer simulations of dragonfly forewings.

A novel genetic algorithm for evolving both uniform and nonuniform cellular automata (CA) to perform user-defined computations is presented. Unlike previous approaches, the CAs evolved here can in general take as their input and their output only a subset of the cells, allowing for the design of CAs that are larger than the number of inputs required by the desired computation. It also provides greater flexibility compared with previous work in terms of the number of possible outputs. We test our algorithm by attempting to evolve both uniform and nonuniform 1D CAs of varying sizes to compute the sum of two 4-bit strings, a computation requiring 8 inputs and 5 outputs. Results demonstrate that while the algorithm is unable to discover solutions using 8-cell CAs, expanding the number of cells beyond the number of inputs enables the autonomous design of 4-bit adders.

A key challenge in evolving control systems for robots using neural networks is training tractability. Evolving monolithic fixed topology neural networks is shown to be intractable with limited supervision in high dimensional search spaces. Common strategies to overcome this limitation are to provide more supervision by encouraging particular solution strategies, manually decomposing the task and segmenting the search space and network. These strategies require a supervisor with domain knowledge and may not be feasible for difficult tasks where novel concepts are required. The alternate strategy is to use self-organized task decomposition to solve difficult tasks with limited supervision. The artificial neural tissue (ANT) approach presented here uses self-organized task decomposition to solve tasks. ANT inspired by neurobiology combines standard neural networks with a novel wireless signaling scheme modeling chemical diffusion of neurotransmitters. These chemicals are used to dynamically activate and inhibit wired network of neurons using a coarse-coding framework. Using only a global fitness function that does not encourage a predefined solution, modular networks of neurons are shown to self-organize and perform task decomposition. This approach solves the sign-following task found to be intractable with conventional fixed and variable topology networks. In this paper, key attributes of the ANT architecture that perform self-organized task decomposition are shown. The architecture is robust and scalable to number of neurons, synaptic connections, and initialization parameters.

A new speciation method for parallel evolutionary computation is presented, designed specifically to handle high-dimensional data. Taking inspiration from the natural sciences, the Phylogenetic Relations Island Speciation Model (PRISM) uses common ancestry and a novel species barcoding system to detect new species and move them to separate islands. Simulation experiments were performed on Multidimensional Knapsack Problems with different fitness landscapes requiring 100-dimensional genomes. PRISM's performance with various parameter settings and on the various landscapes is analyzed and preliminary results show that PRISM can consistently produce optimal or near-optimal solutions, outperforming the standard Genetic Algorithm and Island Model in all the performed experiments.

We present an "artificial neural tissue" (ANT) architecture as a control system for autonomous multirobot tasks. This architecture combines a typical neural-network structure with a coarse-coding strategy that permits specialized areas to develop in the tissue which in turn allows such emergent capabilities as task decomposition. Only a single global fitness function and a set of allowable basis behaviors need be specified. An evolutionary (Darwinian) selection process is used to derive controllers for the task in simulation. This process results in the emergence of novel functionality through the task decomposition of mission goals. ANT-based controllers are shown to exhibit self-organization, employ stigmergy and make use of templates (unlabeled environmental cues). These controllers have been tested on a multirobot resource-collection task in which teams of robots with no explicit supervision can successfully avoid obstacles, explore terrain, locate resource material and collect it in a designated area by using a light beacon for reference and interpreting unlabeled perimeter markings. The issues of scalability and antagonism are addressed

A neural network used in a direct-adaptive control scheme can achieve trajectory tracking of a (highly) flexible joint robot holding an unknown payload without need for many learning repetitions. A modification of the Lyapunov stable nonlinear control method known as backstepping with tuning functions is derived to achieve this. Specifically, the introduction of appropriate weightings of the different tuning-function terms results in high performance. Also, a robust redesign of the tuning function method is presented to account for the uniform approximation (modeling) error of the neural network. This computationally burdensome method is made practical by taking advantage of the efficient structure of the CMAC neural network. Simulations with a (highly) flexible-joint robot show immediate compensation for a payload with performance nearly recovered after five seconds.

We discuss our experiences in integrating a commercial off-the-shelf ground-penetrating radar unit with an all-terrain rover. Straight-line subsurface surveys were generated in a fully autonomous manner using odometry and a simple visual servoing technique. Survey results for various terrains are presented. We discuss the configuration of the integrated system and make recommendations for both Martian and terrestrial applications.

A method of planning paths for formations of mobile robots with nonholonomic constraints is presented. The kinematics equations presented in this paper allow a general geometrical formation of mobile robots to be maintained while the group as a whole travels an arbitrary path. It is possible to represent a formation of mobile robots by a single entity with the same type of nonholonomic constraint as the individual members. Thus, any path-planner or control method may be used with the formation as would be applied to an individual robot. Equations are developed for changing the geometrical formation and hardware results are presented from the Stanford MARS Testbed.

A expansion encoding approach to image classification is presented. Localized principal components or "eigenpaxels" are used as a set of basis functions to represent images. That is, principal-component analysis is applied locally rather than on the entire image. The "eigenpaxels" are statistically determined using a database of the images of interest. Classification based on visual similarity is achieved through the use of a single-layer error-correcting neural network. Expansion encoding and the technique of subsampling are key elements in the processing stages of the eigenpaxel algorithm. Tested using a database of frontal face images consisting of 40 individuals, the algorithm exhibits equivalent performance to other comparable but more cumbersome methods. In addition, the technique is shown to be robust to various types of image noise.

An approach to evolving globally coordinated behaviours in groups of autonomous mobile robots is presented. The control system in each robot is identical and consists of a cellular automaton which serves to arbitrate between a number of fixed basis behaviours. Genetic algorithms search for cellular automata whose arbitration results in success on a predefined task. Heap formation is presented as an example of a task requiring global coordination. Simulation results are provided.

A strategy for locating and grasping a target object in an unknown position using a robotic manipulator equipped with a CCD camera is described. Low-level trajectory and joint control during the grasping operation is handled by the manipulator's conventional motion controller using target-pose data provided by an artificial-neural-network-based vision system. The feature CMAC is a self-organizing neural network that efficiently transforms images of a target into estimates of its location and orientation. The approach emulates biological systems in that it begins with simple image features (e.g., corners) and successively combines them to form more complex features in order to determine object position. Knowledge of camera parameters, camera position and object models is not required since that information is incorporated into the network during a training procedure wherein the target is viewed in a series of known poses. The manipulator is used to generate the training images autonomously. No training of connection weights is required; instead, training serves only to define the network topology which requires just one pass through the training images. Experiments validating the effectiveness of the strategy on an industrial robotic workcell are presented.

An artificial neural network is proposed for the precision control of flexible-joint robots. The training method uses backstepping in an online, direct neuroadaptive scheme in order to guarantee stability. The online weight updates include a learning term that improves performance while maintaining stability. Albus's cerebellar model arithmetic computer algorithm is modified to work for flexible robots by utilizing radial basis functions to deal with the elasticity. The resulting hybrid network is referred to as CMAC-RBF associative memory or CRAM network. Many of the properties of the CMAC for rigid robot control are kept by using CRAM for flexible-joint robots.

These parallel algorithms described are based on a new O(N) solution to the problem. The underlying feature of this O(N) method is a different strategy for decomposition of interbody force which results in a new factorization of mass matrix (M). Specifically, a factorization of inverse of the mass matrix in the form of Schur complement is derived as M/sup -1/=C-D/sup t/A/sup -1/B wherein A, B, and C are block tridiagonal matrices. The new O(N) algorithm is then derived as a recursive implementation of this factorization of M/sup -1/. It is shown that the resulting algorithm is strictly parallel. Strategies for multilevel exploitation of parallelism in the computation are also discussed, resulting in more efficient parallel O(log N) algorithms. The parallel algorithms developed in this paper, in addition to their theoretical significance, are also important from a practical implementation standpoint due to their simple architectural requirements.>

In this paper, two parallel O(log N) algorithms for the computation of manipulator forward dynamics are presented. They are based on a new O(N) algorithm for the problem which is developed from a new factorization of mass matrix M. Specifically, a factorization of the inverse M/sup -1/ in the form of a Schur complement is derived. The new O(N) algorithm is then developed as a recursive implementation of this factorization. It is shown that the resulting algorithm is strictly parallel, that is, it is less efficient than other algorithms for serial computation of the problem. However, to our knowledge, it is the only algorithm that can be parallelized to derive both a time-optimal O(logN) - and processor-optimal - O(N) - parallel algorithm for the problem. A more efficient parallel O(logN) algorithm based on a multilevel exploitation of parallelism is also briefly described. In addition to their theoretical significance, these parallel algorithms allow a practical implementation due to their simple architectural requirements.>

A control concept applicable to the teleoperation of multilink, structurally flexible manipulators, based on commanded velocity of the end-effector, is presented. A velocity-tracking controller is used to effect the desired motion of the manipulator. The controller seeks to minimize the velocity tracking error at the cost of allowing the links to flex. This controller is designed by precomputing gains based on a set of dynamics equations, obtained by linearizing over a space of predetermined geometrical configurations and casting them into state-space form. Appropriate real-time controller gains are determined by interpolating between a subset of those precomputed gains, corresponding to geometrical configurations, in the neighborhood of the desired configuration. The system is augmented with a second-order integrator (for torque smoothing) and further augmented by a servocompensator whose input is the error between the actual end-effector velocity and the reference (command) velocity. Although principally motivated by teleoperation, the controller can also be used in an autonomous mode, which is the focus of this paper.

An experimental study of a control policy for end-effector trajectory tracking of structurally flexible space-based manipulators is presented. The controller employs a fully feedback-driven approach using a series of steady-state linear regulators. An augmented dynamical description involving derivatives of the control inputs is employed to ensure smooth force and/or torque profiles at the a joints. Experiments were performed on Radius, a two-link planar manipulator with flexible links and rotary joints supported on a horizontal table by air pucks. Radius is designed so that its frequencies of vibration are comparable to those of space manipulators. The arm is instrumented with potentiometers for measuring joint angles and rates as well as strain gages for monitoring link deformation. The joints are actuated by DC motors coupled to harmonic drive gear reducers. Experimental results show that the controller is able to track demanding end-effector trajectories very well. These results agree closely with computer simulations.>

A solution procedure for simulation dynamics of elastic multibody systems specifically designed for parallel processing is presented. The method is applicable to open chains with general (rotational and/or translational) interbody constraints. It is based on obtaining an explicit solution for the joint constraint forces by means of iterative techniques. Numerical results for a three-link anthropomorphic flexible-link manipulator are presented. The simulated comparison, on a serial computer, of different parallel iterative schemes indicates that the preconditioned conjugate-gradient methods are computationally most efficient. Moreover, their parallel implementation yields computational complexity that, based on theoretical estimates, is approximately constant with the number of bodies in the chain.>

A recursive algorithm, based on a Newton-Euler formulation, is developed for the solution of the simulation-dynamics problem for a chain of rigid bodies. Arbitrary joint constraints are permitted, that is, joints may allow translational and/or rotational degrees of freedom. The recursive procedure is shown to be identical to that encountered in a discrete-time optimal-control problem. For each relevant quantity in the multibody-dynamics problem, there exists an analog in the context of optimal control. The performance index that is minimized in the control problem is identified as Gibbs' function for the chain of bodies.>

A solution procedure is presented for the simulation dynamics of elastic multibody systems specifically designed for parallel processing. The method is applicable to open chains with general (rotational and/or translational) interbody constraints. It is based on obtaining an explicit solution for the joint constraint forces by means of iterative techniques. Numerical results for a three-link anthropomorphic flexible-link manipulator are presented. Comparison of different parallel iterative schemes indicates that the preconditioned conjugate-gradient method is the most computationally efficient method on a series computer. Moreover, its parallel implementation can potentially outperform the recursive method of analysis.>

A computer simulation procedure for the dynamics of topological chains, using a recursive Newton-Euler formulation, is presented. The bodies of the chain are, in general, elastic and the joints can permit arbitrary (rotational and/or translational) interbody motion. Relative interbody translation, however, is assumed small. As an example, a three-link quasianthropomorphic flexible-link manipulator is studied. The simulation results underscore the importance of modeling structural flexibility.>