Model of Social-Adaptive Navigation of a Mobile Robot Using Reinforcement Learning Methods

Classic trajectory planning algorithms, despite their effectiveness in static environments, demonstrate significant limitations when integrated into a dynamic social environment. The main drawback is the inability to interpret human movement in real time, which leads to unpredictable and potentially dangerous maneuvers. In response to these limitations, reinforcement learning (RL) methods have gained considerable popularity. This paradigm allows an autonomous mobile robot to independently form an optimal behavior strategy through direct interaction with the environment and receiving feedback in the form of rewards or penalties. This study focuses on reinforcement learning methods and social behavior models with the aim of developing safe, effective, and socially adaptive navigation for autonomous mobile robots. This paper presents a model in which the key feature is a comprehensive approach to shaping agent behavior. The proposed model takes into account not only basic tasks, goal achievement, and avoidance of physical obstacles, but also important aspects of social interaction. The scientific novelty of the work lies in the development of a multicomponent reward function that integrates rewards for reaching the target point, avoiding collisions with dynamic and static objects, and purposefully encourages the agent to adhere to socially acceptable norms. In this way, the robot learns not only to avoid people, but to do so in a way that is intuitive and comfortable for them. The ultimate goal of the research is to create a navigation agent that is not only safe, but also socially intelligent. This is a step towards the full integration of autonomous robotic systems into everyday human environments, as successful coexistence requires not only physical safety, but also psychological comfort and intuitive understanding of the robot's behavior.