Development of artificial intelligence self-learning driving algorithm of a mobile robot

Mobile robots have become a significant domain where automation and artificial intelligence technologies are increasingly integrated. These robots, due to their mobility capabilities, are utilized in various application areas, including logistics, agriculture, construction, healthcare, and industrial automation. One of the fundamental components enabling the effective and safe operation of mobile robots is their driving algorithms. Traditional driving algorithms allow robots to operate with pre-programmed instructions for specific scenarios. However, while this approach may be effective for simple and repetitive tasks, it limits the success of robots in complex and dynamic environmental conditions. The integration of artificial intelligence technologies into robots aims to make driving algorithms more adaptive and autonomous by enabling mobile robots to perceive their surroundings and learn from their experiences. This study aims to develop an artificial intelligence-supported self-learning driving algorithm for a mobile robot. The algorithm's adaptive learning mechanisms, compatibility, autonomy, learning speed, and performance are expected to provide valuable contributions to the literature, particularly in terms of real-world applications. To achieve this goal, the study utilized reinforcement learning methods commonly used in the literature, including DDPG, TD3, PPO, SAC, and A2C, to observe and compare their success in complex environments. The complex environments were designed to change in every scene, allowing a single algorithm to be trained and tested under different levels of complexity. According to the findings, SAC achieved 93.6% success, TD3 achieved 86.3%, A2C achieved 89.2%, and PPO achieved 97.6% success. These results are seen as an important step for the future development of mobile robot technology and its effective use in various sectors. In conclusion, this thesis aims to enable mobile robots to operate reliably in different scenarios, thereby enhancing their applicability in diverse fields such as logistics, military operations, search and rescue missions, and domestic assistance tasks. The ability of the algorithms to rapidly adapt to environmental changes has facilitated the development of robots that can function more flexibly and for general-purpose use. This innovative approach, supported by dynamic environments and comprehensive analyses, has improved the autonomous mobility capabilities of mobile robots. It is foreseen that these algorithms will be tested in more complex environments in the future and integrated with various data sources and sensors. This study is expected to provide significant contributions to autonomous systems and artificial intelligence-based robotic applications.