Search

The global community is awaiting the advent of a self-driving vehicle that is safe, reliable, and capable of navigating a diverse range of road conditions and terrains. This requires a lot of research, study, and optimization. Thus, this work focused on implementing, training, and optimizing a convolutional neural network (CNN) model, aiming to predict the steering angle during driving (one of the main issues). The considered dataset comprises images collected inside a car-driving simulator and further processed for augmentation and removal of unimportant details. In addition, an innovative data-balancing process was previously performed. A CNN model was trained with the dataset, conducting a comparison between several different standard optimizers. Moreover, evolutionary optimization was applied to optimize the model's weights as well as the optimizers themselves. Several experiments were performed considering different approaches of genetic algorithms (GAs) along with other optimizers from the state of the art. The obtained results demonstrate that the GA is an effective optimization tool for this problem. [ABSTRACT FROM AUTHOR]

In the last year, thanks to the Ms. Pac-Man Versus Ghosts Competition, the game of Ms. Pac-Man has gained increasing attention from academics in the field of computational intelligence. In this paper, we contribute to this research stream by presenting a simple genetic algorithm with lexicographic ranking (GALR) for the optimization of flocking strategy-based ghost controllers. Flocking strategies are a paradigm for intelligent agents characterized by showing emergent behavior and for having very little computational and memory requirements, making them well suited for commercial applications and mobile devices. In particular, we study empirically the effect of optimizing homogeneous and heterogeneous teams. The computational analysis shows that the flocking strategy-based controllers generated by the proposed GALR outperform the ghost controllers included in the competition framework and some of those presented in the literature. [ABSTRACT FROM PUBLISHER]

Distributed Evolutionary Algorithms are traditionally executed on homogeneous dedicated clusters, despite most scientists have access mainly to networks of heterogeneous nodes (e.g., desktop PCs in a lab). Fitting this kind of algorithms to these environments, so that they can take advantage of their heterogeneity to save running time, is still an open problem. The different computational power of the nodes affects the performance of the algorithm, and tuning or fitting it to each node properly could reduce execution time. Since the distributed Evolutionary Algorithms include a whole range of parameters that influence the performance, this paper proposes a study on the population size. This parameter is one of the most important, since it has a direct relationship with the number of iterations needed to find the solution, as it affects the exploration factor of the algorithm. The aim of this paper consists in validating the following hypothesis: fitting the sub-population size to the computational power of the heterogeneous cluster node can lead to an improvement in running time with respect to the use of the same population size in every node. Two parameter size schemes have been tested, an offline and an online parameter setting, and three problems with different characteristics and computational demands have been used. Results show that setting the population size according to the computational power of each node in the heterogeneous cluster improves the time required to obtain the optimal solution. Meanwhile, the same set of different size values could not improve the running time to reach the optimum in a homogeneous cluster with respect to the same size in all nodes, indicating that the improvement is due to the interaction of the different hardware resources with the algorithm. In addition, a study on the influence of the different population sizes on each stage of the algorithm is presented. This opens a new research line on the fitting (offline or online) of parameters of the distributed Evolutionary Algorithms to the computational power of the devices. [ABSTRACT FROM AUTHOR]

This paper investigates the performance and the results of an evolutionary algorithm (EA) specifically designed for evolving the decision engine of a program (which, in this context, is called bot) that plays Planet Wars. This game, which was chosen for the Google Artificial Intelligence Challenge in 2010, requires the bot to deal with multiple target planets, while achieving a certain degree of adaptability in order to defeat different opponents in different scenarios. The decision engine of the bot is initially based on a set of rules that have been defined after an empirical study, and a genetic algorithm (GA) is used for tuning the set of constants, weights and probabilities that those rules include, and therefore, the general behaviour of the bot. Then, the bot is supplied with the evolved decision engine and the results obtained when competing with other bots (a bot offered by Google as a sparring partner, and a scripted bot with a pre-established behaviour) are thoroughly analysed. The evaluation of the candidate solutions is based on the result of non-deterministic battles (and environmental interactions) against other bots, whose outcome depends on random draws as well as on the opponents' actions. Therefore, the proposed GA is dealing with a noisy fitness function. After analysing the effects of the noisy fitness, we conclude that tackling randomness via repeated combats and reevaluations reduces this effect and makes the GA a highly valuable approach for solving this problem. [ABSTRACT FROM AUTHOR]

• The massive number of possible actions in an RTS game is a challenge for AI. • Using domain knowledge can lead an AI towards promising states. • Small-scale heuristics could help in shaping the space of actions. • Heuristics may be combined to form a wide array of strategies. • Genetic algorithms are able to automatically configure heuristics for optimal play. Real-Time Strategy (RTS) games are well-known for their substantially large combinatorial decision and state spaces, responsible for creating significant challenges for search and machine learning techniques. Exploiting domain knowledge to assist in navigating the expansive decision and state spaces could facilitate the emergence of competitive RTS game-playing agents. Usually, domain knowledge can take the form of expert traces or expert-authored scripts. A script encodes a strategy conceived by a human expert and can be used to steer a search algorithm, such as Monte Carlo Tree Search (MCTS), towards high-value states. However, a script is coarse by nature, meaning that it could be subject to exploitation and poor low-level tactical performance. We propose to perceive scripts as a collection of heuristics that can be parameterized and combined to form a wide array of strategies. The parameterized heuristics mold and filter the decision space in favor of a strategy expressed in terms of parameters. The proposed agent, ParaMCTS , implements several common heuristics and uses Na ï veMCTS to search the downsized decision space; however, it requires a preceding manual parameterization step. A genetic algorithm is proposed for use in an optimization phase that aims to replace manual tuning and find an optimal set of parameters for use by EvoPMCTS , the evolutionary counterpart of ParaMCTS. Experimentation results using the μ RTS testbed show that EvoPMCTS outperforms several state-of-the-art agents across multiple maps of distinct layouts. [ABSTRACT FROM AUTHOR]