Identifying which parts of a scene are actionable and how an agent can physically interact with them.