Kinematic Analysis of an Under-constrained Cable-driven Robot Using Neural Networks

In this paper, three types of neural networks including, multilayer perceptron, radial basis function, and Local Linear Model Trees, are considered based on adaptability with the significant behavior of data to solve the forward and inverse kinematics of the under-constrained cable-driven parallel robot. The practical workspace of the robot is a rectangular cube with dimensions of 360cm × 220c× 160cm. Forward kinematics problem of the under-constrained cable-driven parallel robot is challenging to solve by reason of complexity and nonlinearity of the robot kinematic equations. The typical approach utilizes a numerical method to solve these equations, which are computationally expensive and very time-consuming. Resorting to the neural network approaches results in the feasible orientation, which confirms that all the cables are in tension. The trained neural network gives a feasible inverse kinematic problem solution and the running time of the optimal trained neural network in the control loop, obtained about 12 microseconds, pales in comparison to the sampling time of the control loop that facilitates the real-time control of the robot. Obtained results reveal that the multilayer perceptron has the mean average error order of about 2 millimeters for the position and 0.78 degrees for orientation, which makes a case for precise control in the high-speed motion of the robot.