Development and optimization of realtime inverse kinematics of a delta robot using machine learning

The rapid advancements in industrial automation have significantly increased the demand for high-speed and high-precision robotic systems. Among them, Delta robots have emerged as a widely adopted solution for applications such as pick-and-place operations, assembly lines, and packaging due to their unique parallel kinematic structure. However, the complexity of their inverse kinematics poses a significant challenge for efficient control and trajectory planning. This thesis explores a machine learning-based approach to solving the inverse kinematics problem of Delta robots, aiming to enhance adaptability, precision, and real-time performance. Traditional methods rely on analytically derived models that require extensive calibration and are highly sensitive to mechanical variations and external disturbances. In contrast, this work utilizes artificial neural networks to learn the kinematic relationships directly from data, offering a more flexible and robust alternative. The study involves the acquisition and preprocessing of real-time sensor data, the selection of optimal neural network architecture, hyperparameter tuning, and model training to accurately predict joint configurations based on desired end-effector positions. The proposed approach is validated through simulations and experimental tests, demonstrating accuracy and adaptability compared to conventional techniques. Additionally, the system's robustness to disturbances and mechanical imperfections is analyzed. The focus of this research is to fill the gap between theoretical advancements in machine learning and their practical implementation in industrial robotics. Future work will focus on further optimizing the neural network model, implementing real-time control mechanisms based on energy optimization and optimal path following and exploring hybrid approaches combining analytical models with data-driven learning for enhanced performance.