Sandeep Gulati

Neutral learning of constrained nonlinear transformations

Two issues that are fundamental to developing autonomous intelligent robots, namely rudimentary learning capability and dexterous manipulation, are examined. A powerful neural learning formalism is introduced for addressing a large class of nonlinear mapping problems, including redundant manipulator inverse kinematics, commonly encountered during the design of real-time adaptive control mechanisms. Artificial neural networks with terminal attractor dynamics are used. The rapid network convergence resulting from the infinite local stability of these attractors allows the development of fast neural learning algorithms. Approaches to manipulator inverse kinematics are reviewed, the neurodynamics model is discussed, and the neural learning algorithm is presented.<<ETX>>

A neural network based identification of environments models for compliant control of space robots

Many space robotic systems would be required to operate in uncertain or even unknown environments. The problem of identifying such environment for compliance control is considered. In particular, neural networks are used for identifying environments that a robot establishes contact with. Both function approximation and parameter identification (with fixed nonlinear structure and unknown parameters) results are presented. The environment model structure considered is relevant to two space applications: cooperative execution of tasks by robots and astronauts, and sample acquisition during planetary exploration. Compliant motion experiments have been performed with a robotic arm, placed in contact with a single-degree-of-freedom electromechanical environment. In the experiments, desired contact forces are computed using a neural network, given a desired motion trajectory. Results of the control experiments performed on robot hardware are described and discussed. >

Application of adjoint operators to neural learning

A new methodology for neural learning of nonlinear mappings is presented. It exploits the concept of adjoint operators to enable a fast global computation of the networks response to perturbations in all system parameters.

Terminal slider control of robot systems

Many robotic systems would, in the future, be required to operate in environments that are highly unstructured (with varying dynamical properties) and active (possessing means of self-actuation). Although a significant volume of results exist in model-based, robust and adaptive control literature, many issues pertinent to the stabilization of contact interactions with unpredictable environments remain unresolved, especially in dealing with large magnitude and high frequency parametric uncertainties. The primary intent of this paper is nonlinear control synthesis for robotic operations in unstructured environments. We introduce the notion oftime constrained terminal convergence for controlled systems, and propose an approach to nonlinear control synthesis based upon a new class of sliding modes, denotedterminal sliders. Terminal controllers that enforce finite convergence to equilibrium are synthesized for an example nonlinear system (with and without parametric uncertainties). Improved performance is demonstrated through the elimination of high frequency control switching, employed previously for robustness to parametric uncertainties [2]. The dependence of terminal slider stability upon the rate of change of uncertainties over the sliding surface, rather than the magnitude of the uncertainty itself, results in improved control robustness. Improved reliability is demonstrated through the elimination ofinterpolation regions [2]. Finally, improved (guaranteed) precision is argued for through an analysis of steady state behavior.

Parallelizing the cascade-correlation algorithm using Time Warp

Abstract This paper discusses the method in which the cascade-correlation algorithm was parallelzznd in such a way that it could be run using the Time Warp Operating System (TWOS). TWOS is a special-purpose operating system designed to run parallel discrete event simulations with maximum efficiency on parallel or distributed computers. The parallelization process is described, and a formula is derived for the maximum possible speedup using this technique. For the benchmark used, a speedup of 8 was obtained while running on 10 nodes of a BBN GP1000 parallel computer, indicating that this approach is a useful way of parallelizing cascade-correlation.

The pebble-crunching model for fault-tolerant load balancing in hypercube ensembles

We propose a graph-theoretic, receiver-initiated, distributed protocol for dynamic load balancing protocol in large-scale hypercube ensembles. Using attributed hypergraphs as the primary data structure for constraint modelling and dynamic optimisation, we consider systems running precedence-constrained heterogeneous tasks. Fault Tolerance is ensured by incorporating a dynamic integrity check for the decision nodes and their subsequent re-election if needed. Simulation studies are used to analyse the algorithm performance and correctness

Self-organizing neuromorphic architecture for manipulator inverse kinematics

We describe an efficient neuromorphic formulation to accurately solve the inverse kinematics problem for redundant manipulators, thereby enabling development of enhanced anthropomorphic capability and dexterity. Our approach involves a dynamical learning procedure based on a novel formalism in neural network theory: the concept of “terminal” attractors, that are shown to correspond to solutions of the nonlinear neural dynamics with infinite local stability. Topographically mapped terminal attractors are then used to define a neural network whose synaptic elements can rapidly encapture the inverse kinematics transformations using a priori generated examples and, subsequently generalize to compute the joint-space coordinates required to achieve arbitrary end-effector configurations. Unlike prior neuromorphic implementations, this technique can also systematically exploit redundancy to optimize kinematic criteria, e.g. torque optimization, manipulability etc. and is scalable to configurations of practical interest. Simulations on 3-DOF and 7-DOF redundant manipulators, are used to validate our theoretical framework and illustrate its computational efficacy.

Control of space robots in uncertain environments using neural networks

Many space robotic systems would be required to operate in uncertain or even unknown environments. In this paper, the problem of identifying such environments for compliance control is considered. In particular, neural networks are used for identifying environments that a robot establishes contact with, with fixed nonlinear structure and unknown parameters. Compliant motion experiments have been performed with a robotic arm, where desired contact forces are computed using a neural network.<<ETX>>

Smelting Networks For Real Time Cooperative Planning In The Presence Of Uncertainties

This paper discusses the applicability and limitations of the conventional knowledge representations and Al paradigms in designing algorithms for multi-sensor fusion for the uncertainty prone multiple target tracking (MTT) problem. A knowledge-based framework for solution to this problem necessitates developing mechansims for efficient online knowledge acquisition and exploration of alternate knowledge representations as the existing structures prove inadequate in this context. In this paper we describe the use of Smelting Networks for knowledge acquisition, target state representation, resolution of sensor ambiguities and environmental uncertainties. These networks are drawn from a biological cooperative phenomenon and used to enrapture situational constraints, determine track-to-target associations and avoid generation of redundant associations for trajectory estimation.