Affordance

What an object or environment lets you do with it — a handle can be pulled, a button can be pressed.

The concept of affordance was introduced by American psychologist James J. Gibson in 1966 and developed fully in his 1979 book The Ecological Approach to Visual Perception, describing the possibilities for action that an environment offers to an animal. It is relational: the same chair affords sitting to a person and affords jumping-onto to a cat. Don Norman brought the term into interaction design in 1988. In robotics, affordance answers “how can this object be interacted with, and from where” — a cup's handle affords grasping, for instance. Common approaches mark interactable regions on an image or point cloud, known as affordance detection, or, as in SayCan, use a value function to estimate whether a skill can succeed right now, which keeps a large model's plans limited to what the robot can actually do.

ExampleLooking at a mug, an affordance model marks the handle as a “graspable” region and the rim as a “pourable-into” region, and the robot decides where to approach it from based on that.

Related

Affordance Detection · SayCan · Grasping · Articulated Object Manipulation · Embodied Perception · Language Grounding

Sources

Affordance - Wikipedia
SayCan: Do As I Can, Not As I Say: Grounding Language in Robotic Affordances