Robot Grip Robustness Prediction using a Hybrid Deep Learning Approach

Humans have been using robots now for over a decade. With the latest developments in Electrical and Computer Engineering robots now can perform very critical tasks with high degrees of precision which was not possible in earlier times. Today robots perform various tasks such as industrial welding, space exploration, deep sea exploration, human surgery, bio-mechatronics etc. Now robot manufacturers have a wide variety of options to choose from while designing and developing their robots, from the mechanical parts to the software capabilities of the robot. One of the most common application of robots is to perform pick-and-place operations. The aim of this study is to predict the robustness in gripper robots using Deep Learning models so that the robot can be designed according to their robustness in safety critical environments. This is an important area as the robustness of the grip is a critical feature that must be ensured if the robot is to work with pick and place operations. The type of gripper to be used and its effectiveness determines the true value and efficiency of the robot. We have performed the analysis on shadow dataset obtained through Shadow Robot. The robot used in this study has a 3-finger gripper which is used to pick and place a ball. The proposed approach using an Artificial Neural Network (ANN) to extract features and Linear regression to predict the robustness factor of the grip. The model provides us an R-Squared score of 0.86 and an RMSE value of 18.12 which is sufficiently good and can help determine efficiently the robustness of the grip of the robot.