Automated fixture designs using reinforcement learning

The design of industrial fixtures for quality inspections requires high accuracy due to the precise measurement of complex components. Current approaches are largely not fully automated, resulting in disadvantages such as long processing times, the need for potential manual adjustments, and limited scalability for complex fixtures. One promising method to address these limitations is the development of a reinforcement learning (RL) based model. Consequently, this paper presents the automation of the fixture design process using RL. In the proposed modeling, the fixture design problem is formulated as a multi-agent RL algorithm, with agents handling part selection and part placement under consideration of multiple optimization criteria. Furthermore, the paper addresses modeling-related challenges, focusing on the required state information, the artificial neural network structure of the agents, and the action spaces.