The paper presents a study on the evolution of neural networks using genetic algorithms in the context of the RoboCup Keepaway Testbed. The goal was to develop a learning algorithm for autonomous agents in a simplified soccer environment. The Keepaway Testbed was scaled down to a simulated world where a single agent must locate a goal. This environment provided a realistic setting for testing learning algorithms. A simulated world was created with walls, an agent, and a goal. The agent uses sensors to determine its position, angle, and the location of the goal, and effectors to turn and move. The task was to navigate the world and locate the goal. The researchers used a feed-forward neural network to learn the task. Traditional back-propagation was considered, but genetic algorithms were chosen for their ability to optimize neural network weights. The genetic algorithm was used to evolve the weights of the neural network, with the fitness of each individual determined by the number of steps required to find the goal. The neural network had one input layer, two hidden layers, and one output layer. Inputs included the agent's position, the goal's position, and the agent's heading. Outputs determined the agent's actions, such as turning or moving. The genetic algorithm was designed with a population of 200 individuals, each represented as a string of real numbers. The algorithm used stochastic universal sampling for selection, an elitist scheme to preserve the best individual, and a combination of recombination and mutation to introduce diversity. The algorithm was run for 1500 iterations, with the best individual achieving a fitness of 0.16. The learning curve showed that the genetic algorithm could learn the weights for the neural network, and the average fitness typically peaked early. Despite the low fitness values, the agent demonstrated intelligent behavior in navigating to the goal. The study concluded that the approach was novel but had been previously documented by Peter Stone. The research was halted, but the team plans to revisit the problem in the future.The paper presents a study on the evolution of neural networks using genetic algorithms in the context of the RoboCup Keepaway Testbed. The goal was to develop a learning algorithm for autonomous agents in a simplified soccer environment. The Keepaway Testbed was scaled down to a simulated world where a single agent must locate a goal. This environment provided a realistic setting for testing learning algorithms. A simulated world was created with walls, an agent, and a goal. The agent uses sensors to determine its position, angle, and the location of the goal, and effectors to turn and move. The task was to navigate the world and locate the goal. The researchers used a feed-forward neural network to learn the task. Traditional back-propagation was considered, but genetic algorithms were chosen for their ability to optimize neural network weights. The genetic algorithm was used to evolve the weights of the neural network, with the fitness of each individual determined by the number of steps required to find the goal. The neural network had one input layer, two hidden layers, and one output layer. Inputs included the agent's position, the goal's position, and the agent's heading. Outputs determined the agent's actions, such as turning or moving. The genetic algorithm was designed with a population of 200 individuals, each represented as a string of real numbers. The algorithm used stochastic universal sampling for selection, an elitist scheme to preserve the best individual, and a combination of recombination and mutation to introduce diversity. The algorithm was run for 1500 iterations, with the best individual achieving a fitness of 0.16. The learning curve showed that the genetic algorithm could learn the weights for the neural network, and the average fitness typically peaked early. Despite the low fitness values, the agent demonstrated intelligent behavior in navigating to the goal. The study concluded that the approach was novel but had been previously documented by Peter Stone. The research was halted, but the team plans to revisit the problem in the future.