Implementing a Semi-Autonomous Robot for Ball Tracking in the University of Jember ABU Robot Based on Deep Learning

The development of robotics technology is growing rapidly. Indonesian Robot Contest (Kontes Robot Indonesia - KRI) is a form of competition for developing robotics in Indonesia. This research develops a semi-autonomous robot that can perform robot missions semi-autonomously. The semi-autonomous ABU robot is designed with a ball object detection system using the Convolutional Neural Network method of yolov5 algorithm consisting of 213 layers. Before the implementation of the detection, a model is obtained after the training process, which requires 2,500 training image data and 353 testing data with a hyperparameter of 100 epochs and 24 batches. The process produces a model with a mAP value of 0.850. The detection system can achieve a detection accuracy value of 78.3% - 100% in offline and real-time testing with differences in object distance and light intensity when detecting objects. Then, in the object tracking system accompanied by robot movements that have been tested 10 times with different angular ranges of objects, the average travel time is 2.27 seconds to reach the object's center point.