Generación de comportamientos en robots autónomos mediante una arquitectura cognitiva híbrida

Abstract

In the field of artificial intelligence and robotics, one enduring objective is to create autonomous robots capable of performing complex tasks in dynamic environments. In these environments, robots not only need to react in predefined or deterministic scenarios but also learn and adapt in real-time, mirroring a cognitive flexibility akin to human intelligence. Achieving this autonomy entails developing cognitive architectures that seamlessly integrate reactive, deliberative, and emergent capabilities. Over the last few decades, various approaches—deliberative systems, three-layer architectures, reactive models, and emergent systems—have been explored, all aiming to generate intelligent behaviors in autonomous robots. Thus, this research work focuses on the design, development, deployment and evaluation of a hybrid cognitive architecture to generate, control, plan, and monitor behaviors in autonomous robots. This architecture amalgamates reactive, deliberative, and emergent components, aiming to enhance adaptability in dynamic environments and make intelligent real-time decisions, thereby improving autonomy and performance. In the current landscape of robotics research, the development of hybrid architectures has emerged as a critical path to address limitations seen in purely reactive or purely deliberative architectures. Deliberative systems excel in symbolic reasoning and planning but face challenges when dealing with the dynamic and unpredictable nature of real-world environments. Conversely, reactive systems excel in tasks requiring rapid responses to realtime sensor stimuli but lack high-level reasoning and planning capabilities. The combination of deliberative, reactive, and emergent capacities, alongside layered organization, results in the architecture presented in this work. The systematic literature review undertaken has focused on recent research articles addressing this topic, resulting in insights into recent cognitive architectures, behavior-generating tools, methods, and metrics used to evaluate these architectures. Tools such as state machines, behavior trees, PDDL, symbolic planners, knowledge graphs, and various types of emergent components have been identified. Different methods for evaluating cognitive architectures, notably experiments involving human interaction, and metrics such as execution time, distance traveled, and performance, have also been highlighted. In conclusion, a hybrid cognitive architecture integrated into ROS 2 for autonomous robots, termed MERLIN2, is presented. MERLIN2 comprises a deliberative system based on a knowledge base and a symbolic planner, a behavior system composed of reactive components, and several emergent components. It addresses core cognitive aspects like perception, action selection, memory, learning, reasoning, and explainability. Finally, the experimentation presented showcases the architecture as a valid solution for autonomous robots.