Orientation Prediction in Robotics: A Study of Trigonometric Decomposition Methods Across Synthetic and Real-World Datasets

Orientation prediction is a critical task for robotics as it enables robots to understand and interact with their environment more effectively. By accurately determining an object's position and orientation, robots can perform a range of complex tasks. This in turn will advance smart manufacturing facilities to achieve higher levels of automation, increase efficiency, and enable more flexible production systems. Hence, we present a comparative study of shallow regression models, integration strategies, and trigonometric encoding schemes for planar orientation prediction in robotics, using synthetic and real-world datasets. Results demonstrate that XGBoost 1.7, combined with vector integration and quadrant encoding, achieves the best balance of accuracy, robustness to angular boundary discontinuities, and computational efficiency, significantly outperforming alternative approaches in real-world scenarios.