This work presents a reinforcement learning–based approach to target-oriented dexterous grasping using a multi-fingered robotic hand integrated with a redundant industrial manipulator. Unlike many existing methods, the proposed framework does not rely on visual input during policy training. Instead, grasp configurations, including the end-effector pose and finger joint positions, are generated by an external grasp synthesis module and used as target states for goal-conditioned reinforcement learning. This separation between grasp generation and execution simplifies training and reduces computational requirements. Domain randomization is applied to object pose and robot configuration to improve robustness and generalization in simulation. The approach enables functional grasping, allowing o

Дауд Г. (науч. рук. Борисов И.И.) Target-oriented dexterous grasping with a multi-fingered robotic hand integrated on a redundant industrial manipulator // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2026]. URL: https://kmu.itmo.ru/digests/article/17528