Saleh Zein-Sabatto

Dynamic multiobjective optimization of war resource allocation using adaptive genetic algorithms

Genetic algorithms (GA) are often well suited for multiobjective optimization problems. The major objective of this research is to optimize the war resource allocations of sorties, for a given war scenario, using genetic algorithms. The war is simulated using THUNDER software. THUNDER software is a stochastic, two-sided, analytical simulation of campaign-level military operations. The simulation is subject to internal unknown noises similar to real war cases. Due to these noises and discreteness in the simulation, as well as in real wars, an adaptive GA approach has been applied to solve this multiobjective optimization problem. Transforming this multiobjective optimization problem to a form suitable for direct implementation of GA was a major accomplishment of this research. A suitable fitness function was chosen after careful research and testing on the GA. Furthermore, the GA parameters were adaptively set to yield smoother and faster fitness convergence. Two fuzzy logic mechanisms were used to adapt the GA parameters. In the first mechanism, the mutation and crossover rates were changed adaptively. In the second mechanism, the fitness function coefficients are changed dynamically in each run. Testing results showed that the adaptive GA outperforms the conventional GA search in this multiobjective optimization problem and was effectively able to allocate forces for war scenarios.

Multiple path planning for a group of mobile robot in a 2-D environment using genetic algorithms

Considers the development and implementation of an approach using genetic algorithms for finding optimum paths for a group of mobile robots located at arbitrary starting positions to a given number of targets in a known multi-obstacle environment. The factors considered for finding the optimum paths for the group of mobile robots are the location and size of obstacles in the environment. The environment is first converted into a grid map. Each grid is assigned a weight value, which indicates the level of confidence for the robots to move on that grid. The grid map contains information about the positions of the robots, the targets, the obstacle locations and sizes. Two genetic algorithms modules have been developed to find the optimum paths for the mobile robots. The first genetic algorithm module takes information about the environment from the grid map and finds an obstacle-free optimum path for each mobile robot to each target. The second genetic algorithm module finds the best combination of mobile robots to move to a given number of targets.

Genetic algorithms applied to real time multiobjective optimization problems

Genetic algorithms (GAs) are often well-suited for multi-objective optimization problems. In this work, multiple objectives pertaining to the THUNDER software (a very large military campaign simulation model) were used to optimize the war results obtained from the software. It is a stochastic, two-sided, analytical Monte-Carlo simulation of military operations. The simulation is subject to internal unknown noises. Due to these noises and to the discreteness in the simulation program, a GA approach has been applied to this multi-objective optimization problem. This method is capable of searching for multiple solutions concurrently in a single run. Transforming this problem to a form that is suitable for the direct implementation of GA was the major challenge that was achieved. Three different kinds of fitness assignment methods were implemented, and the best one was chosen. The THUNDER software may be considered as a black box, since very little information about its internal dynamics was known. The problem with the THUNDER software is its expensive running time. In order to optimize the time involved with the THUNDER software, autocorrelation techniques were used to reduce the number of THUNDER runs. Furthermore, the GA parameters were set optimally to yield smoother and faster fitness convergence. From these results, the GA was shown to perform well for this multi-objective optimization problem and was effectively able to allocate force power for the THUNDER software.

Vehicle identifications using acoustic sensing

Increase in the complexity of battlefield fought in urban areas has brought a high demand for efficient techniques for vehicles detection, classification, identification and tracking in areas of interest. The demand for such efficient techniques is due to complexity of the environment and to sensor limitations. Multi-sensor data fusion can be used for vehicle identification in practical applications such as battlefield surveillance. Multi-sensor data fusion provides significant advantages over single sensor data source. The purpose of this research is to design and implement data fusion software for vehicle surveillance applications using a distributed network of acoustic sensors. The data fusion software will be used in vehicles traffic monitoring in a battlefield and events detection and tracking in secured areas. Implementation and testing results of the developed software are obtained from real data collect from civilian and military vehicles.

Localization strategies for large-scale airborne deployed wireless sensors

Localization is the process of finding the geometric locations of wireless sensor nodes according to some real or virtual coordinate system. It is an important task when direct measurements of the wireless sensor locations are not available. From the various techniques evolved in localizing sensor nodes, one approach is to use the received signal strength to predict the location of unknown sensing devices. In this paper, passive localization algorithms are developed, presented and tested. The algorithms perform region-based localization of stationary wireless sensors with respect to a frame of reference using received signal strength of the sensors. The reported work is conducted in two phases, theoretical development then simulation and hardware testing. In the first phase, localization algorithms were developed to predict the location of wireless sensor nodes. We categorized localization of sensors in three different classes. In class-I, localization is done for sensors that are in the communication range of at least three head nodes. In class -II, localization is done for sensors in the communication range of two head nodes, and in class-III, localization is done for sensors that are in the communication range of only one head node. In the second phase, the three different categories were tested by simulation then using hardware. A test-bed was established using the crossbow (MICAz) hardware and used to measure the sensors transmission signal strength. Then, the localization software provided with these signal strength as input to predict the location of each wireless sensor nodes. The algorithm developments, the simulation and hardware preliminary test results of the localization algorithms are presented in this paper.

Evolutionary approach to multi-objective problems using adaptive genetic algorithms

The paper describes an adaptive genetic algorithm used to achieve multi-objectives such as minimizing the territory losses and maximizing enemy air losses by finding the optimum distribution of aircraft fighting in a war scenario simulated by the THUNDER software. The adaptive genetic algorithm changes the mutation and crossover adaptively to provide fast convergence to the optimum possible solutions. According to the population of the fitness values obtained for each generation, three distribution properties (the mean, the variance and the best fitness value) are determined and used as input to a fuzzy-logic system for modifying the mutation and crossover rates to obtain the individuals of the next generation. This enables fast and smooth convergence to the best possible solutions.

Modeling human-robot interaction for intelligent mobile robotics

The focus of this paper is the design of a system for human-robot interaction that allows the robot(s) to interact with people in modes that are common to them. The results are a designed architecture for a system to support human-robot interaction. The structure includes a monitoring agent for detecting the presence of people, an interaction agent to handle choosing robot behaviors that are used for interacting, both socially and for task completion, and a capability agent which is responsible for the robots abilities and actions.

Human-robot interaction over the Internet

Describes two Java-based Internet control mechanisms, manual control and autonomous navigation, for a mobile robot to allow users to explore the robots environment. In the manual control, a user sends control commands to the robot over the Internet to make the robot explore its environment. In the autonomous navigation mode, the user only needs to specify the target point and the rest of the task is taken care of by the robot. In both cases, the real time video images, the robot trajectory, and a status report of the robot are sent to the user over the Internet. In addition, sonar readings are transmitted over the Internet in the manual control to provide more detailed information to the user. The mobile robot was successfully controlled from several locations in the USA.

Fuzzy controller design using genetic algorithms

A methodology for combining genetic algorithms (GAs) with fuzzy controllers to create genetic/fuzzy controllers is presented. Using GAs, optimal or near optimal fuzzy rules and membership functions can be designed without a human operators experience or a control engineers knowledge, although such information can be used for the initial design. This genetic/fuzzy approach involves searching the encoded fuzzy rule and membership function parameter spaces using a fitness function that is defined in terms of a system performance criterion. We demonstrate this approach in an application where a GA adapts the fuzzy rules and membership functions of a fuzzy controller for a tracking system in real-time. The generalization ability of this tracking system is demonstrated by training it only on a step input, freezing its adaptable parameters, and then showing that it can accurately track other types of input signals.

A neural network-based text independent voice recognition system

A text-independent voice recognition experiment was conducted using an artificial neural network. The speech data were collected from three different speakers uttering thirteen different words. Each word was repeated ten times. The speech data were then pre-processed for signal conditioning. A total of 12 feature parameters were obtained from Cepstral coefficients via a linear predictive coding (LPC). These feature parameters then served as inputs to the neural network for speaker classification. A standard two-layer feedforward neural network was trained to identify different feature sets associated with the corresponding speakers. The network was tested for the remaining unseen words in text-independent mode. The results were very promising with a voice recognition accuracy of more than 90%. The success rate could be increased by adding more utterances from each speaker.<<ETX>>