Aprendizado sensório-motor em robôs cognitivos utilizando modelo CST-CONAIM

• Ano de defesa: 2021
• Autor(a) principal: Rossi, Leonardo de Lellis
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Artificial intelligence; Cognition; Automação; Cognição; Engenharia elétrica; Inteligência artificial
• Link de acesso: http://hdl.handle.net/11449/214316

Cognitive architectures, as the CONAIM model (Conscious Attention-Based Integrated Model) (SIMõES, 2015; SIMõES; COLOMBINI; RIBEIRO, 2016; SIMOES; COLOMBINI; RIBEIRO, 2017; COLOMBINI; SIMõES; RIBEIRO, 2017), suggest that a way to obtain consciousness in machines is the use of several factors inspired by those present in the human cognitive system, such as memories, reasoning, planning, decision making, learning, motivation, and attention. However, implementing this type of system has shown to be complex, requiring computational tools with a high degree of specialization or implemented in very limited situations. A specialized platform for building cognitive architectures, CST (Cognitive Systems Toolkit), stands out in this scenario (GUDWIN et al., 2013; PARAENSE et al., 2016). It contains cognitive structures called codelets and memory objects, capable of supporting the modules provided in the CONAIM model, as well as their possible evolutions. Recent works of our research group implemented CONAIM and reinforcement learning (RL) modules with the CST (REGATTIERI; COLOMBINI, 2018; BERTO, 2020; BERTO et al., 2020a; BERTO et al., 2020b) with a single procedural learning mechanism that, without external modifications, demonstrates the incremental learning process in the first three sensorimotor substages in Jean Piaget’s Theory (PIAGET, 1952). The validation of the development and learning process of this system will be carried out through a collection of cognitive experiments that can be used to study the new model and its computational and cognitive impacts (BERTO, 2020). Thus, the present work seeks to extend this investigation by implementing the learner attentive-cognitive agent and its validation through some of the proposed cognitive experiments using mobile robots in simulated environments. The results obtained show the possibility of evaluating the learning level of a robot through incremental cognitive experiments.