Mauricio A. Dias

Hardware/Software co-design using artificial neural network and evolutionaryy computing

Hardware/Software co-design is an important problem nowadays and is involved in many hardware research and development fields. Hardware/software partitioning problem is one of the most important questions on co-design that defines how parts of a hardware/software system should be implemented on which a fast and good solution is essential. This paper proposes a new way on hardware/software partitioning problem solving using an artificial neural network trained with resilient back propagation algorithm and feed-forward architecture.

Swarm intelligence and the quest to solve a garbage and recycling collection problem

This work focuses on the application of Swarm Intelligence to a problem of garbage and recycling collection using a swarm of robots. Computational algorithms inspired by nature, such as Particle Swarm Optimization (PSO) and Ant Colony Optimization, have been successfully applied to a range of optimization problems. Our idea is to train a number of robots to interact with each other, attempting to simulate the way a collective of animals behave, as a single cognitive entity. What we have achieved is a swarm of robots that interacts like a swarm of insects, cooperating with each other accurately and efficiently. We describe two different PSO topologies implemented, showing the obtained results, a comparative evaluation, and an explanation of the rationale behind the choices of topologies that enhanced the PSO algorithm. Moreover, we describe and implement an Ant Colony Optimization (ACO) approach that presents an unusual grid implementation of a robot physical simulation. Hence, generating new concepts and discussions regarding the necessary modifications for the algorithm towards an improved performance. The ACO is then compared to the PSO results in order to choose the best algorithm to solve the proposed problem.

Genetic Algorithm Applied to Robotic Squad Coordination

The main goal of this paper is to describe the modeling, implementation and evaluation of the Genetic Algorithms (GA) efficiency when applied to robotic group formation and coordination. The robotic task in this paper is performed over a natural disaster, simulated as a forest fire. The robot squad mission is to surround the fire and avoid fires propagation. Experiments have been made with different chromosome models and several parameters variation. This paper describes all performed experiments detailing all sets of parameters, including positive and negative results. The simulations results showed that with an adequate set of parameters is possible to get satisfactory strategic positions for a multi-robotic systems operation; furthermore, this GA solution can be applied on similar activities.

A driving assistance system for navigation in urban environments

Mobile robot navigation in urban environments is a very complex task. As no single sensor is capable to deal with these situations by itself, sensor fusion techniques are required to allow safe navigation in this type of environment. This paper proposes an approach to combine different sensors in order to assist a driver in a cooperative manner. An adaptive attention zone in front of the vehicle is defined and the driver is notified about obstacles presence, identifying dangerous situations. Experiments using a commercial vehicle loaded with GPS and a LIDAR sensor have been performed in real environments in order to evaluate proposed approach.

Automatic Generation of LUTs for Hardware Neural Networks

A common solution for activation function design in HNN is the LUT representation that has acceptable area constraints and usually achieves acceptable execution time. Tools and techniques for automatic generation of LUTs have been developed but have some problems as high complexity, high dependency of external tools or languages and the optimization techniques that are not efficient enough for critical applications. In order to solve some of the presented problems this paper describes a method for automatic LUT generation that overcomes some deficiencies of existing methods using a simple approach. This method was evaluated considering two different robotic systems that hereafter will be implemented in hardware and whose networks have a different behaviour using different domains of the same activation function. Proposed method achieved good results that are directly related to how networks define the activation function domain.

Hardware/Software Co-design Implementation of On-Chip Backpropagation

Artificial neural networks are a parallel, fault tolerant, robust solution for computational tasks such as associative memories, pattern recognition and function approximation. There are many proposed implementations for artificial neural networks and networks learning algorithms both in hardware and software. Hardware implementation of learning algorithms are a computational challenge because some constraints as maximum number of neurons and layers, training time, precision, and data representation are difficult to be optimized together. This paper describes a hardware/software co-design implementation of the error-back propagation algorithm on multi-layer perceptron networks. Different types of processors, with different hardware features and goals, were created and the results were analyzed considering mentioned constraints. The results present a hardware/software co-design that allows a large number of neurons and layers, that maintains initial precision without restrictions on data representation. Platform limitations resulted in high execution times but solutions to this problem are also proposed. So the developed hardware proved to be a good alternative considering current hardware implementations of training algorithms and also the mentioned requirements.

HW/SW Co-design Architecture for Evolutionary Robotics

Evolutionary algorithms are very common techniques used in computational intelligence and robotics field applications. Some algorithms need a large amount of memory and processing power, making them difficult to implement into embedded systems. In this work a profile-based approach is proposed and applied in an evolutionary algorithm with some characteristics that allow it’s use on embedded systems and robotics: the micro-GA. The main goal is to implement a new hardware-software co-design architecture for this genetic algorithm with better execution time than algorithms implemented in software (using general purpose hardware solutions). The presented results show a comparison between different code sign implementations and discussion about new architecture advantages.

Fixed-Point Neural Network Ensembles for Visual Navigation

Visual navigation is an important research field in robotics because of the low cost and the high performance that is usually achieved by visual navigation systems. Pixel classification as a road pixel or a non-road pixel is a task that can be well performed by Artificial Neural Networks. In the case of real-time instances of the image classification problem, as when applied to autonomous vehicles navigation, it is interesting to achieve the best possible execution time. Hardware implementations of these systems can achieve fast execution times but the floating-point implementation of Neural Networks are commonly complex and resource intensive. This work presents the implementation and analysis of a fixed-point Neural Network Ensemble for image classification. The system is composed by six fixed-point Neural Networks verified with cross-validation technique, using some proposed voting schemes and analyzed considering the execution time, precision, memory consumption and accuracy for hardware implementation. The results show that the fixed-point implementation is faster, consumes less memory and has an acceptable precision compared to the floating-point implementation. This fact suggests that the fixed point implementation should be used in systems that need a fast execution time. Some questions about ensembles and voting have to be reviewed for fixed-point Neural Network Ensembles.

Hardware/Software Co-design for Image Cross-Correlation

Cross-correlation is an important image processing algorithm for template matching widely used on computer vision based systems. This work follows a profile-based hardware/software co-design method to develop an architecture for normalized cross-correlation coefficient calculus using Nios II soft-processor. Results present comparisons between general purpose processor implementation and different customized softprocessor implementation considering execution time and the influence of image and sub-image size. Nios II soft-processors configured with floating-point hardware acceleration achieved a 8.31 speedup.

Genetic Algorithms and Artificial Neural Networks to Combinational Circuit Generation on Reconfigurable Hardware

Operating in critical environments is an extremely desired feature for fault-tolerant embedded systems. In addition, due to design test and validation complexity of these systems, faster and easier development methods are needed. Evolvable Hardware (EHW) is a development technique that, using reconfigurable hardware, builds systems that reconfiguration part is under the control of an Evolutionary Algorithm. Reconfigurable hardware allows EHW to change its own hardware structure adapting itself to task and/or environment changes. Evolvable part of these systems can also be implemented using Artificial Neural Networks (ANNs). This research work presents results and comparisons between Genetic Algorithm (GA) and ANN implementations that receive combinational circuits’ truth-tables as input and searches the minimum circuit respecting this input truth-table. GA improved for this work’s EHW structure achieve good execution time for tested tables and ANN modeling presents some non-desired characteristics with bad results.