Robotic Arm Handling Based on Real-time Gender Recognition Using Convolutional Neural Networks

This paper presents the development of a system for controlling a robotic arm to deliver an object depending on the gender identity (male or female) of a human recognized in front of the robot to demonstrate essential gender identification-based applications for physical Human-Robot Interaction. For this, we developed a convolutional neural network-based model for identifying genders. With the recognition result, the control of the robotic arm with six degrees of freedom was implemented using a Jetson Nano embedded computer, OpenCV, ROS, and TensorFlow libraries. The developed gender identification model achieved a 96.5% of accuracy and a loss of 3.5% during training and validation using a gender database composed of 50K gender images. The final real-time prototype obtained a 98.2% accuracy and a margin of error of 1.8% during testing. This proof of concept indicates that more complex applications based on the gender of the user could also be developed in the future.