This thesis investigates trajectory correction for robotic manipulation of deformable materials. In industrial settings, robots often struggle to interact accurately with non-rigid objects such as flexible plastic lids due to their unpredictable shapes and poses. To address this challenge, two distinct perception-driven systems were developed and evaluated. The first system is a rule-based machine vision approach utilizing classical image processing techniques to detect object orientation and adjust predefined robot trajectory points in real time. The second system is an AI-based model using a convolutional neural network (CNN) trained on annotated image data to predict corrected dot positions directly from visual input. The systems were tested under three key scenarios: normal object placement, rotation along the z-axis, and partial or full shape deformation. The results show that the machine vision system performs reliably in ideal and occluded conditions due to its explicit geometric logic, while the AI model achieves faster inference and strong generalization under rotational changes. However, the AI model lacks internal logic for occlusion handling, sometimes resulting in invalid predictions. In conclusion, both systems successfully correct robotic trajectories, with trade-offs between explainability and adaptability. The rule-based system offers robustness in safety-critical cases, whereas the AI model provides a scalable, learning-driven solution that benefits from richer datasets and deeper deformation awareness.