Real-world context-aware grasping scenes