affordance

In robotics, an affordance is a property or feature of an object in the environment that enables a specific action or interaction by the robot. When a robot's vision system detects a handle on a door, that handle affords grasping; when it sees a flat surface at the right height, that surface affords placement of an object. Affordances are relational: they exist not in the object alone, but in the pairing of the object with the robot's capabilities and constraints. A small ledge affords purchase for a gripper with 10 mm jaws but not one with 20 mm minimum clearance.

The concept emerged from ecological psychology but has become essential in modern robotic manipulation and navigation. Rather than programming explicit rules for every action, engineers build systems that learn or detect affordances: visual features that signal how an object can be used. A robot learning from human demonstration might identify that cylindrical shapes with textured surfaces afford gripping at certain diameters, or that downward-facing surfaces afford sliding motion. The robot associates visual and tactile cues with successful manipulation outcomes.

Affordances in perception and control

In practice, a robot must infer affordances from sensor data, whether cameras, depth sensors, force feedback, or tactile arrays. A pushing task requires surfaces that afford sliding without excessive friction; a stacking task requires surfaces that afford stable contact and weight distribution. Deep learning systems trained on manipulation data can predict affordance maps, typically spatial feature matrices overlaid on camera images, that indicate where and how interaction is possible. These maps drive motion planning and gripper configuration.

Affordance mismatch is a common failure mode. A robot trained to recognize handles on cylindrical objects may fail on oval or D-shaped handles. A gripper calibrated for smooth plastic may slip on oily metal. Environmental variation, sensor noise, and wear on mechanical fingers all degrade affordance detection. This is why robust systems combine multiple modality inputs and include force or torque sensing to confirm that an intended affordance actually exists before committing to a forceful action.

The term itself comes from the design theory of James Gibson, who used it to describe the possibilities for action that an environment presents to an organism. In robotics, affordances bridge the gap between perception and action: they are the interface between what the robot observes and what it can do. Understanding affordances is now central to sim-to-real transfer, where robots trained in simulation must recognize that affordances in the physical world may differ subtly from those in the simulated model.