APLICAÇÃO DE APRENDIZAGEM POR REFORÇO ADAPTATIVO PARA O DESVIO DE OBSTÁCULOS EM AGENTES VIRTUAIS

• DOI: https://doi.org/10.22533/at.ed.8089122527115

• Palavras-chave: Aprendizado de máquina. Aprendizagem por reforço. Sistemas adaptativos. Inteligência Artificial. Autonomia computacional.

• Keywords: Machine learning. Reinforcement learning. Adaptive systems. Artificial Intelligence. Autonomous agents.

• Abstract: Machine learning is one of the fundamental pillars of contemporary Artificial Intelligence, enabling computational systems to infer patterns, extract relationships, and make autonomous decisions from data. Among its classical approaches—supervised learning, unsupervised learning, and reinforcement learning—the latter stands out for its agent–environment interaction paradigm, in which an agent learns optimal action policies through trial, error, and accumulated rewards. In this context, adaptive reinforcement learning plays a crucial role, as it enables the agent to adjust its behaviour in dynamic, non-stationary, or uncertain environments. By incorporating adaptive mechanisms, the agent shifts from a static policy to a continuously refined decision strategy, enhancing responsiveness to structural variations and stochastic perturbations. This adaptive capability makes the method particularly suitable for mobile robotics, intelligent control, autonomous vehicles, advanced manufacturing, gaming, and complex industrial processes. Furthermore, adaptive reinforcement techniques increase model robustness, generalization capacity, and resilience. Approaches such as dynamic adjustment of the exploration–exploitation balance, meta-learning, real-time hyperparameter optimization, and integration with deep learning architectures significantly improve agent performance and autonomy. Therefore, adaptive reinforcement learning is an essential component for the development of intelligent systems capable of operating efficiently, safely, and reliably in complex and evolving environments.