테이블 가장자리 식기의 안정적 파지를 위한 DQN기반 밀기 경로 탐색

Abstract

At the table edge, it may be difficult to grasp the target dish depending on the positions of surrounding dishes. This paper proposes the DQN (Deep Q-Network) based method for generating pushing paths to grasp dishes stably at the table edge. This study developed an algorithm that combines reinforcement learning and search algorithms to effectively push target dish to the position where it can be picked up, in situations where it is difficult to directly grasp the dish because of the surrounding dishes. To validate the performance and effectiveness of our method, we experiment in various simulation environments, and analyze the results.