This research aims to enhance humanoid robot soccer Ball Tracking, Goal Alignment, and Robot avoidance tasks using YOLO-NAS. The study followed a three-stage approach involving model engineering, which involves model training, code integration, and testing by comparing it with YOLO-v8 and YOLOv7. We measured the mAP (Mean Avegara Precision) and the speed of the detection of each model. Descriptive and Friedman techniques were employed to interpret testing results. In the ball tracking task, YOLO-NAS achieved a success rate of 53.3 compared to YOLOv7 with 68.3. In the goal alignment task, YOLO-NAS achieved the highest success rate of 91.7. In the Robot Avoidance task, YOLO-NAS, the same as YOLOv8, 100 nailed the test. These findings suggest that YOLO-NAS performs exceptionally well in the goal-alignment task but does not excel in two other tasks related to humanoid robot soccer. © 2024 Department of Agribusiness, Universitas Muhammadiyah Yogyakarta. All rights reserved.