Glossary
Affordance
What a robot can actually do in a given situation, as distinct from what it might want to do.
An affordance is a possible action enabled by an object or environment. The term comes from psychology (Gibson): a door affords being opened, a cup affords being held, a button affords being pressed. In robotics, affordance is what filters a planner's wish list down to actions that are physically possible.
In modern stacks, affordance models often pair with language-model planners. SayCan, for example, lets the LLM propose plausible next actions ("pick up the cup") and an affordance model scores each ("can this robot, in this scene, do this?"). The score gates the action.
Affordance-prediction models often share a backbone with the policy itself, so the system has a unified view of what it can and can't do.