Autonomous robotic assembly process based on hierarchical reinforcement learning

Presently, the assembly process is planned by a trained individual, which takes expertise and time to plan. Since the flexibility and optimality of the assembly plan are dependent on the expert s knowledge and creativity, expertise is an important parameter in developing the assembly plan. As a result, researchers have explored the use of intelligent methods to design the assembly process. To automate the robotic assembly process, Reinforcement Learning (RL) can be used as a powerful approach to handle complicated mechanical assembly tasks without explicit programming. This paper proposes a novel approach to automate the assembly process that is inspired by Hierarchical Reinforcement Learning (HRL). This method aims to tackle the assembly problem by breaking it into multiple sub-problems, teaching each sub-problem with an appropriate algorithm, and then merging them. The use of this technique is promising for solving problems that can be broken into smaller independent ones, which are called sub-problems. To evaluate and illustrate the capability of the proposed method, a case study is conducted using Soft Actor-Critic (SAC) algorithm as the main learning algorithm to assemble the components of a gearbox, and obtained results are discussed.