Robotic Grasping

The field studies how a robot selects contact points and gripper poses, then controls its hand to acquire and manipulate objects safely. Common approaches include geometric grasp sampling from RGB-D images or point clouds, learned grasp policies and diffusion or flow-based generators, and language- or affordance-guided planning for task-specific contacts. Tactile and force feedback, compliant gripper designs, and simulation-to-real training help systems adjust to slip, object properties, and changing contact during execution. A central challenge is making grasps physically reliable and efficient under clutter, partial observations, unknown friction, and different robot embodiments, while benchmarks increasingly test real-world execution as well as predicted grasp quality.