Design of a self-driving mini-robot for indoor navigation using evolutionary artificial intelligence algorithms.

Autonomous systems have many applications as space exploration, assembling tasks, and household maintenance. These systems have to be able to adapt to a wide variety of tasks. Reinforcement learning approaches and evolutionary approaches are two major fields with adaptability characteristics. The Evolutionary artificial intelligence (NEAT) algorithm is a combination of these two approaches, and, for this reason, in this work, it is used for autonomous navigation purposes in conjunction with color recognition approaches. The NEAT algorithm has been widely used in video game applications and in simulated environments. The present work attempts to extend the uses of the NEAT by using it in an image processing and robotic application introducing an adapted NEAT algorithm, called NEAT for self-driving with color recognition (NEAT-SDCR). The main aim of the present work is to discover if NEAT-SDCR, performs well in a real indoor environment. For a first approach, the assigned task is simple; a robot is trained to follow green objects. The training is done in a simulated environment; then it is tested in a real robot. The contributions of this work are the implementation of this system in a real robot, the design of a fitness function for the problem, and the improvement in general of the accuracy of the NEAT algorithm by the use of a new reproduction method and by using an incremental approach.

• Päätekijä: González Núñez, Joseph Ricardo (author)
• Aineistotyyppi: bachelorThesis
• Kieli: eng
• Julkaistu: 2020
• Aiheet: Autonomous system; Reinforcement learning; Evolutionary approach; NEAT; Sistemas autónomos; Aprendizaje por refuerzo; Enfoques evolutivos
• Linkit: http://repositorio.yachaytech.edu.ec/handle/123456789/184