Detection of kiwifruits during harvest is essential for maximizing yield and reducing loss. Hand harvesting is time-consuming and susceptible to human error, whereas automated systems are challenged by issues like high canopy density, changing illumination, and the brittle nature of kiwifruit. Existing object detection algorithms are lacking in offering a tradeoff between speed, accuracy, and flexibility in such challenging environments. This research overcomes such hurdles through improved real-time object detection in Kiwi fruit picking via the YOLO11x-mod model and sophisticated hyperparameter optimization. This research contributes to precision agriculture technology innovation via the upgrade of the performance of robotic systems in agricultural use, thereby enhancing productivity and minimizing reliance on human labour. Nine hundred twenty-eight photos and 22,750 labelled kiwifruits datasets were utilized with rotation, brightness change, and Gaussian noise as data augmentation methods to introduce diversity and avoid overfitting. YOLOv8, YOLOv10, and various YOLOv11 variants were tried out, and the optimization of YOLO11x-mod was done through hyperparameter tuning in terms of anchor box resizing and learning rate scheduling. The YOLO11x-mod model excelled with a precision of 0.8176, a Recall of 0.8397, and mAP50 of 0.8619. Its ability to have superior detection of kiwifruits under adverse conditions, including foliage density and different lighting conditions, makes it the model of choice when used in automated harvesting systems, indicating its excellence.