Self-location Algorithm for the Strategic Movement of Humanoid Robots

This work presents an algorithm that enables one or more humanoid robots to position themselves within a specific area by identifying the goal and determining the distance between the robot and the objective. The main objective is to facilitate the operation of the system in soccer-playing robots. To achieve this, two premises are established. Firstly, in the hypothetical scenario where two robots from the same team compete for the ball, the algorithm determines which robot is closer and makes the decision of who will go for the objective, thereby avoiding collisions among the robots. Subsequently, the robot will move towards the target area, i.e., the goal. Secondly, the hypothetical case is considered in which each robot belongs to a different team, and both will attempt to reach the ball before the other. In both situations, the robot performs the identification of the ball and the other robot, whether friend or foe, in order to determine the distances between the robots and the ball by applying Deep Learning algorithms. For object identification, YOLOv3, a classic model for object identification that uses convolutional neural networks, was retrained. The computer vision is implemented by a RealSense camera which uses stereoscopic vision to calculate the depth of the objects. The robots utilize a self-localization and strategy algorithm, which is executed by each robot’s command system. There is no direct communication between the robots in either scenario. However, a wireless communication system between the robot and the computer is required, with Wi-Fi as the initial choice. The algorithm’s performance will be evaluated using measures such as accuracy, precision, and sensitivity, employing tools like the confusion matrix and the Receiver Operating Characteristic (ROC) curve.