High-Precision Peg-in-Hole Assembly with Flexible Components Based on Deep Reinforcement Learning

Abstract

The lateral thrust device is a typical high-pressure sealed cavity structure with dual O-rings. Because the O-ring is easily damaged during the assembly process, the product quality is unqualified. To achieve high-precision assembly for this structure, this paper proposes a reinforcement learning assembly research method based on O-ring simulation. First, a simulation study of the damage mechanism during O-ring assembly is conducted using finite element software to obtain damage data under different deformation conditions. Secondly, deep reinforcement learning is used to plan the assembly path, resulting in high-precision assembly paths for the inner and outer cylinder under different initial poses. Experimental results demonstrate that the above method not only effectively solves the problem that the O-ring is easily damaged but also provides a novel, efficient, and practical assembly technique for similar high-precision assemblies.