This study aims to address the challenges of the automated inspection of concrete bridge piers, as existing climbing robots often encounter constraints related to their adhesion methods. Furthermore, achieving consistent omnidirectional movement on these large vertical structures is a complex control task that evaluates traditional estimation and control techniques.
This study addresses these limitations by presenting an innovative climbing robot featuring a distinctive force-controlled gripping mechanism for robust, adaptable adhesion and Mecanum wheels for full omnidirectional movement. To achieve accurate control in complex scenarios, this paper presents two methodologies: a Long Short-Term Memory Random Forest (LSTM-RF) model that integrates data from multiple Inertial Measurement Units (IMU) for reliable attitude estimation, and a Deep Q-Network Tube Model Predictive Control (DQN-Tube MPC) framework for motion control.
The controller’s remarkable tracking accuracy, with a Root Mean Square Error (RMSE) of 1.17 cm for the Zr-axis position, 1.17° for Roll, 0.78° for Pitch and 1.725° for Yaw angle.
The proposed robotic system offers an effective and dependable solution for the automated inspection of bridge piers.
