Estudo do desenvolvimento de um veículo autônomo com viabilidade econômica e inteligência artificial

Abstract

Autonomous vehicles have proving to be the future of locomotion. Being able to go from point A to point B leaving the navigation of the vehicle in the hands of technology, allowing it to make the decisions for itself. Achieving this required several advances in electronics, computing, automation and robotics. One type of vehicle that emerged long before the application of technologies to make cars autonomous were AGVs (Automated Guided Vehicles). This type of vehicle is used in various types of industry assisting in the production and storage system. This work aims to study the constructive aspects of this type of vehicle and electronic prototyping platforms and sensors that can be applied to develop a low cost vehicle coupled with the implementation of artificial intelligence through artificial neural networks and reinforcement learning. Research shows that building an AGV that meets the objective requirements is feasible. A virtual vehicle was simulated in three different scenarios with different obstacles and different objectives for testing the artificial neural network to analyze the vehicle's behavior before a real implementation in a future prototype. The simulation results show that the complexity of the environment where the agent is inserted, ie the arrangement of obstacles, influences the amount of learning cycles. With consolidated learning, the next steps for future implementations are building a small-scale prototype and later an AGV.