Non-Prehensile Manipulation Actions and Visual 6D Pose Estimation for Fruit Grasping Based on Tactile Sensing

Robotic manipulation in cluttered environments is one of the challenges roboticists are currently facing. When objects to handle delicate fresh fruits, grasping even more challenging. Detecting and localizing fruits with accuracy necessary grasp them very difficult due large variability aspect dimensions each item. This paper proposes a solution that exploits state-of-the-art neural network novel enhanced 6D pose estimation method integrates depth map output. Even an accurate localization, suitable force avoid slippage damage at same time another challenge. work solves this issue by resorting controller based on tactile sensing. Depending specific application scenario, fruit might be impossible without colliding other or fruits. Therefore, non-prehensile action here proposed push items hindering detected fruit. The pushing from initial location target performed model predictive taking into account unavoidable delay perception computing pipeline robotic system. Experiments real demonstrate overall approach allows robot successfully apples various situations.