Enhanced Deep Reinforcement Learning for Robotic Manipulation: Tackling Dynamic Weight in Noodle Grasping Task

Handling food items with dynamic weight changes over time, which alter physical properties such as shape, size, and weight, poses significant challenges, particularly when precise output weight is required.This study introduces an enhanced deep reinforcement learning framework for robotic manipulation, focusing on the task of spaghetti grasping.Building on prior research, we propose a data augmentation strategy that simulates diverse environmental conditions, including variations in image observations and the physical properties of spaghetti, to improve models.The model is validated using metrics such as grasp success rate, average grasp time, and generalization score under varying environmental conditions.This work advances the robustness of robotic models in previously unseen environments.