Navigating mobile robots through complex and dynamic environments poses a significant challenge in the field of robotics. While Deep Reinforcement Learning (DRL) has advanced robotic control, persistent issues such as suboptimal policies, reward function design complexity, and the inability of Deep Neural Networks (DNNs) to generalize across competing objectives (e.g., speed vs. safety) limit performance. Manual reward shaping, which requires balancing penalties and incentives for diverse tasks, often leads to unintended behaviors or oversimplified policies. In this work, we address these limitations by proposing a multi-task reinforcement learning framework for a ground mobile robot, enabling simultaneous optimization of goal-oriented navigation and obstacle avoidance through a structured

Бархум М., Динь Н. (науч. рук. Борисов О.И., Пыркин А.А.) Multi-task oriented deep reinforcement learning for robot navigation // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2025]. URL: https://kmu.itmo.ru/digests/article/14811