Gilles Labonté

Comparison of Parallel Genetic Algorithm and Particle Swarm Optimization for Real-Time UAV Path Planning

The development of autonomous unmanned aerial vehicles (UAVs) is of high interest to many governmental and military organizations around the world. An essential aspect of UAV autonomy is the ability for automatic path planning. In this paper, we use the genetic algorithm (GA) and the particle swarm optimization algorithm (PSO) to cope with the complexity of the problem and compute feasible and quasi-optimal trajectories for fixed wing UAVs in a complex 3D environment, while considering the dynamic properties of the vehicle. The characteristics of the optimal path are represented in the form of a multiobjective cost function that we developed. The paths produced are composed of line segments, circular arcs and vertical helices. We reduce the execution time of our solutions by using the “single-program, multiple-data” parallel programming paradigm and we achieve real-time performance on standard commercial off-the-shelf multicore CPUs. After achieving a quasi-linear speedup of 7.3 on 8 cores and an execution time of 10 s for both algorithms, we conclude that by using a parallel implementation on standard multicore CPUs, real-time path planning for UAVs is possible. Moreover, our rigorous comparison of the two algorithms shows, with statistical significance, that the GA produces superior trajectories to the PSO.

FPGA Implementation of Genetic Algorithm for UAV Real-Time Path Planning

The main objective of an Unmanned-Aerial-Vehicle (UAV) is to provide an operator with services from its payload. Currently, to get these UAV services, one extra human operator is required to navigate the UAV. Many techniques have been investigated to increase UAV navigation autonomy at the Path Planning level. The most challenging aspect of this task is the re-planning requirement, which comes from the fact that UAVs are called upon to fly in unknown environments. One technique that out performs the others in path planning is the Genetic Algorithm (GA) method because of its capacity to explore the solution space while preserving the best solutions already found. However, because the GA tends to be slow due to its iterative process that involves many candidate solutions, the approach has not been actively pursued for real time systems. This paper presents the research that we have done to improve the GA computation time in order to obtain a path planning generator that can recompile a path in real-time, as unforeseen events are met by the UAV. The paper details how we achieved parallelism with a Field Programmable Gate Array (FPGA) implementation of the GA. Our FPGA implementation not only results in an excellent autonomous path planner, but it also provides the design foundations of a hardware chip that could be added to an UAV platform.

An algorithm for the construction of matrix representationsfor finitely presented non-commutative algebras

Let a finite presentation be given for an associative, in general non-commutative algebra E, with identity, over a field. We study an algorithm for the construction, from this presentation, of linear, i.e. matrix, representations of this algebra. A set of vector constraints which is given as part of the initial data, determines which particular representation of E is produced. This construction problem for the algebra is solved through a reduction of it to the muchsimpler problem of constructing a Grobner basis for a left module. The price paid for this simplification is that the latter is then infinitely presented. Convergence of the algorithm is proven for all cases where the representation to be found is finite dimensional; which is always the case, for example, when E is finite. Examples are provided, some of which illustrate the close relationship that exists between this method and the Todd-Coxeter coset-enumeration method for group theory.

Probabilistic neural networks for infrared imaging target discrimination

The next generation of infrared imaging trackers and seekers will allow for the implementation of more smarter tracking algorithms, able to keep a positive lock on a targeted aircraft in the presence of countermeasures. Pattern recognition algorithms will be able to select targets based on features extracted from all possible targets images. Artificial neural networks provide an important class of such algorithms. In particular, probabilistic neural networks perform almost as optimal Bayesian classifiers, by approximating the probability density functions of the features of the objects. Furthermore, these neural networks generate an output that indicates the confidence it has in its answer. We have evaluated the the possibility of integrating such neural networks in an infrared imaging seeker emulator, devised by the Defense Research and Development establishment at Valcartier. We describe the characteristics extracted from the images and define translation invariant features from these. We give a basis for the selection of which features to use as input for the neural network. We build the network and test it on some real data. Results are shown, which indicate a remarkable efficiency of over 98% correct recognition. For most of the images on which the neural network makes its mistakes, even a human expert would probably have been mistaken. We build a reduced version of this network, with 82% fewer neurons, and only a 0.6% less precision. Such a neural network could well be used in a real time system because its computing time on a normal PC gives a rate of over 5,300 patterns per second.

On a Neural Network that Performs an Enhanced Nearest-Neighbour Matching

Abstract: We review some of the main methods of solving the image matching problem in Particle-Tracking Velocimetry (PTV). This is a technique of Experimental Fluid Dynamics for determining the velocity fields of moving fluids. This problem is a two-dimensional random-points matching problem that condtitutes a prototypal problem, analogous to the one-dimensional matching problem for Julesz [1] random-dot stereograms. Our study deals with a particular method of solution, namely the neural network algorithm proposed by Labonté [2,3]. Our interest in this neural network comes from the fact that it has been shown to outperform the best matching methods in PTV, and the belief that it is actually a method applicable to many other instances of the correspondence problem. We obtain many new results concerning the nature of this algorithm, the main one of which consists in showing how this neural network functions as an enhancer for nearest-neighbour particle image matching. We calculate its complexity, and produce two different types of learning curves for it. We exhibit the fact that the RMS error of the neural network decreases at least exponentially with the number of cycles of the neural network. The neural network constructs a Self-Organised Map (SOM), which corresponds to distorting back the two photos until they merge into a single photo. We explain how this distortion is driven, under the network dynamics, by the few good nearest-neighbours (sometimes as few as 20%) that exist initially. These are able to pull with them the neighboring images, toward their matching partners. We report the results of measuremnts that corroborate our analysis of this process.

Canadian artic Sovereignty: Local intervention by flocking UAVs

The importance of local intervention capability for the assertion of Canadian Sovereignty in the Northwest Passage is recognized. However, Canada lacks the ability to deploy, on demand, assets to search a wide area for rescue or surveillance purposes in the North. This fact motivated our investigation of the feasibility of a rapid intervention system based on a carrier-scouts design in which a number of unmanned aerial vehicles (UAVs) would be transported, air launched and recovered by a carrier aircraft. These UAVs would have the ability to self-organize in formations that correspond to the task at hand. When searching for a target, they would fly in a linear pattern so that the search area swept per hour and the probability of detecting the target would be considerably increased. A 1973 report by the Tactical Combat Aircraft Programs of the Boeing Aerospace Company for the US Air Force and a 2007 thesis by Chalamont indicate that airborne launch and recovery of many UAVs from a carrier aircraft is feasible and requires only already existing technology. We propose here a solution to the remaining problem of managing simultaneously the many UAVs that are required by the vastness of the areas to be surveyed, with a minimum number of human controllers and communications. Namely, we present algorithms for the self-organization of the UAVs in the required formations. These allow for surveillance operations during which close-up images would be acquired of activities in a region of interest, and searching an area for assets in distress and providing a visual presence for such. We reach the conclusion that our proposed local intervention system with flocking UAVs is feasible and would provide a valuable asset for asserting Canadian Sovereignty in the North.

Infrared target-flare discrimination using a ZISC hardware neural network

The last generation of infrared imaging aircraft seekers and trackers uses pattern recognition algorithms to find and keep a lock on an aircraft in the presence of decoy flares. These algorithms identify targets, based on the features of the various objects in the missile’s field of view. Because modern both aircrafts and missiles fly faster than sound, speed of operation of the target identifier is critical. In this article, we propose a target recognition system that respects this time constraint. It is based on an artificial neural network implemented in hardware, as a set of parallel processors on a commercially available silicon chip called a ZISC, for zero instruction set computer. This chip would be integrated in the infrared missile seeker and tracker. We describe the characteristics of the images that the image processing module of this seeker and tracker extracts from the infrared video frames and show how to construct from these translation and rotation invariant features that can be used as input to the neural network. We determine the individual discriminating power of these features by constructing their histograms, which allows us to eliminate some as not being useful for our purpose. Finally, by testing our system on real data, we show that it has a 90% success rate in aircraft-flare identification, and a processing time that during this time, the aircrafts and missiles will have traveled only a few millimeters. Most of the images on which the neural network makes its mistakes are seen to be hard to recognize even by a human expert.

Parallel Algorithm on Graphics Processing Unit for Harmonic Minimization in Multilevel Inverters

This paper presents the implementation details of a parallel algorithm on graphics processing units (GPUs) to compute the optimal switching angles for the harmonic minimization in multilevel inverters with unequal dc voltage sources. Two algorithms, the Newton-Raphson method and the bisection method, and three different parallel implementations are investigated. Both algorithms considered have a low time complexity and offer a superior converging rate allowing for the real-time control of inverters with a very large number of levels. By exploiting the massively parallel architecture of GPUs, the execution time of the program is reduced significantly. The proposed parallel implementation offers a maximum speedup of 534× compared with a sequential execution on CPU, and allows for the calculation of the optimal switching angles for inverters with up to 1000 dc sources in less than 16.4 μs.

Network Parallel Computing for SOM Neural Networks

This paper presents the implementation of a Self-Organizing Map neural network on a distributed computer composed of heterogeneous workstations in order to speedup the computation and to make better use of available computing resources. The developed application is characterized by its dynamic load balancing schemes and its Java implementation. To demonstrate the performance gained, the above mentioned distributed version of a SOM was put to test by realizing neural networks of various sizes. For relatively large neural networks, the distributed version of the SOM is seen to provide linear speedups.

A hardware neural network for target tracking

The Zero Instruction Set Computer (ZISC) is an integrated circuit devised by IBM to realize a restricted Coulomb energy neural network. In our application, it functions as a parallel computer that calculates the correlation coefficients between an input pattern and patterns stored in its neurons. We explored the possibility of using the ZISC in a target tracking system by devising algorithms to take advantage of the ZISCs parallelism and testing them on real video sequences. Our experiments indicate that the ZISC does improve appreciably the computing time compared to a sequential version of the algorithm.