Bottom line: "Vision is simply not suited to the nature of the problem: grasping tasks are a matter of contact and forces, which cannot be monitored by vision.

Bottom line: "Vision is simply not suited to the nature of the problem: grasping tasks are a matter of contact and forces, which cannot be monitored by vision. At best, vision can inform the robot about finger configurations that are likely to succeed, but in the end a robot needs tactile information."