Robotics affordance models

Lens adapted to this topic: What established robotics research generally does

Mainstream robotics research treats affordances as useful computational abstractions linking perceived object or environmental properties to actions and effects. It does not require one formalism: researchers use Bayesian learning, symbolic ontologies, visual prediction, simulation, language grounding, and intermediate representations. The practical emphasis is on enabling robots to generalize beyond fixed action sequences while handling uncertainty and task context.

• Affordance models are widely used to connect objects, actions, effects, and robot capabilities for interaction and planning.
• A formal ontology can propose candidate affordances through logical inference and validate them with physics-based simulation.
• Probabilistic models learned through robot-object interaction can represent uncertainty and support action-effect prediction, object selection, and planning.
• Recent work uses affordances as lightweight intermediate representations that guide robot pose and manipulation policies and may improve generalization.
• Visual affordance research includes grasp detection, classification, segmentation, and hand-object interaction, but differing formulations and datasets create reproducibility problems.

Lens adapted to this topic: Challenges to standard computational assumptions

Critical and ecological perspectives question whether computational models have settled what an affordance is. One line argues that affordances are relational action possibilities grounded in agent-environment dynamics, making purely representational accounts problematic. Another argues that environmental data alone cannot yield conventions of object use, and that robust affordance understanding would require semantics, physics, combinations of objects, and social knowledge. These critiques challenge claims that visual or symbolic labels alone capture the full concept.

• There is considerable disagreement about the nature and definition of affordances, including whether they are invariant objects of perception.
• A Gibsonian account treats affordances as relations between an agent and its environment; critics of representationalism argue that detached internal descriptions can separate perception from action.
• Environmental data by itself may not determine conventions of human object use, which are part of the semantics often attached to affordances.
• Object-affordance prediction may require combining semantics with physics and knowledge of object combinations, not merely detecting visual features.
• Different computational redefinitions of visual affordances can make comparisons between methods and benchmarks unreliable.