Deep Learning-Based Deflection Correction and End-Point Control of Heavy-Duty Vertical Single-Link Flexible Manipulators

Abstract

Flexible link manipulators (FLM) have gained significant importance due to their applications in lightweight robots, energy-efficient systems, and humanoid robots. In this paper, we propose a novel approach to modeling and controlling a single-link flexible manipulator (SLFM). First, a Vertical Single-Link Manipulator (VSLFM) is modeled, taking gravity effects into account, using the Hamilton principle. Data on the link's properties, payload mass, and target angle are used as features to predict the deflection as output, based on numerical analysis of partial differential equation model. For the first time, a Deep Neural Network (DNN) is proposed and trained offline to predict the static deflection of the payload in a set of VSLFMs, using the data obtained from the numerical analysis. Utilizing the predicted deflection, a modified PID controller is developed to control the arc position of the payload. This controller does not require deflection feedback, making it ideal for industrial applications with limited sensors. The method is experimentally validated using an SLFM, which is validated based on a ground truth system consisting of inertial measurement unit-based sensor network.