What a camera watching the scene can't tell you
Vision-based perception, a camera observing the workspace, is excellent at telling a robot where an object is, roughly what shape it has, and where to position the gripper before contact. What it cannot reliably tell the robot is what is happening at the moment contact is made: how much force the fingers are actually applying, whether that force is enough to prevent slipping, or whether the object has begun to slide out of grip. These are properties of the physical interaction between gripper and object, and they are largely invisible to a camera watching from a distance, especially once the object is partially or fully occluded by the gripper's own fingers.
This is the gap that force and tactile sensing built into the gripper itself is designed to close. Force sensors, often placed in the fingers or at the wrist, measure how much force is being applied at the point of contact, letting the controller adjust grip strength directly rather than relying on an open-loop setting decided before contact. Tactile sensors go further, measuring the detailed pattern and distribution of contact across a finger surface, which can reveal not just how hard the gripper is squeezing but how the contact is distributed and how it's changing over time, information a single force sensor summarized as one number cannot provide.