Force Estimation in a 6-DOF UR5 Robot Using Machine Learning Algorithms for Precise Force Control

Robot arms inherently exhibit nonlinear dynamics, which makes force control a challenging task. In each posture of the robot arm, the end-effector deflection varies depending on the applied force. This means that even under the same force input, the resulting deflection may differ due to the configuration-dependent stiffness and dynamics of the system. Such nonlinear behavior complicates the simultaneous control of both force and position, requiring careful balancing and advanced control strategies. In this study, an improved force estimation approach was implemented using the deflection and position information of a robot arm under external force. In this regard, force estimation for a 6-degree-of-freedom UR5 robot is performed using Random Forest, Artificial Neural Networks (ANN), and Gradient Boosting Machine Learning (ML) algorithms. The industrial robot joint angle values and the deflection data in the x, y, and z directions of the end effector are used as the inputs to the ML model and the weight values applied to the end-effector used as the target values. The proposed machine learning algorithm estimates the weight values applied to the end effector using the industrial robot's joint position and its deflection data. The ANN approach outperforms results obtained by other investigated approaches in estimating force value across different robot postures. Based on the results obtained, this method will be used in the next study to define the robot's stiffness in force and position control.