Cable-driven parallel robot control based on a neural network-aided vision system

Cable-driven parallel robots are extensively used in industrial applications, including painting robots, rehabilitation devices, 3D printing systems, and more, due to their high speed and cost-efficiency. However, developing precise models for trajectory planning and controlling these flexible systems takes time and effort. In this study, a novel vision-based position estimation method was proposed that is both accurate and cost-effective. The method employs two stationary smartphone cameras placed 155 cm in front of the robot’s workspace. By implementing ArUco markers, the perspective transformation matrix and the pixel-per-meter were calculated. The robot’s position was then determined using a color detection technique in the HSV color space. This approach enabled the approximate estimation of the robot’s position in real-world coordinates. To enhance the precision of the position estimation, a Multi-Layer Perceptron neural network was trained using a dataset of 99 randomly selected x and y coordinates on a vertical plane. At first, the root mean squared error between the estimated and actual positions was 1.00 cm. However, incorporating the Multi-Layer Perceptron model decreased the error significantly to around 0.46 cm. This vision-based position estimation system was then used as feedback to create a closed-loop system. As the robot navigates from one location to another, the developed system accurately estimates its position and updates the robot’s knowledge of its current location. As a result, the robot is able to plan its trajectory accordingly and effectively stay on its intended path.