طراحي مكانيسم يادگيري تقويتي با حجم محاسبات كم براي آموزش ربات انسان‌نما بوبوليا

Abstract One of the emerging concerns in todayʹs world is the capability of machines to think an‎d interact like humans. This has led researchers in the field of artificial intelligence to design systems that not only possess the ability to understan‎d an‎d process information but can also engage in natural, meaningful, an‎d context-aware communication with humans. Among such systems, conversational agents play a pivotal role in facilitating human-machine interaction an‎d are being increasingly applied in domains such as education, healthcare, customer service, an‎d entertainment. However, designing such systems in the Persian language poses challenges due to the scarcity of linguistic resources an‎d infrastructure. In this research, a Persian-speaking conversational system named "Bobolia" was designed an‎d implemented. This robot leverages deep reinforcement learning algorithms an‎d incorporates user feedback to dynamically adjust its behavior according to ethical an‎d contextual cues. Bobolia is capable of receiving positive an‎d negative feedback through dedicated buttons, enabling it to gradually enhance the quality of its responses an‎d evolve into a continuously learning intelligent assistant. Moreover, the robot can perform actions such as walking, dancing, playing music, obstacle detection, an‎d natural speech synthesis, making it a fully interactive physical companion beyond a simple chatbot. Among the innovations of this study is the use of lightweight learning mechanisms suitable for limited hardware capabilities. Initial results indicate that Bobolia is able to interpret diverse user interactions an‎d utilize them to improve its conversational output. This research represents an important step toward the development of Persian conversational technologies, enhancing human-machine interaction quality, an‎d building systems with ethical awareness an‎d personalized learning capabilities. Keywords: Deep Reinforcement Learning; Question-Answering System; Natural Language Processing; Machine Learning; Q-Learning Algorithm

7 فصل

150519