• NEW!! A Comprehensive Survey + Formalization Attempt

  • To afford or not to afford: A new formalization of affordances towards affordance-based robot control. , Erol Sahin, Maya Cakmak, Mehmet R. Dogar, Emre Ugur, and Gokturk Ucoluk, Adaptive Behavior (2007). web page
  • Definitions - Descriptions - Insight
  1. What Is an Affordance? By: Jones, Keith S.. Ecological Psychology, 2003, Vol. 15 Issue 2, p107-114
  2. Affordances as Properties of the Animal-Environment System. By: Stoffregen, Thomas A.. Ecological Psychology, 2003, Vol. 15 Issue 2, p115-134
  3. An Outline of a Theory of Affordances. By: Chemero, Anthony. Ecological Psychology, 2003, Vol. 15 Issue 2, p181-195
  4. Affordances: Four Points of Debate. By: Michaels, Claire F.. Ecological Psychology, 2003, Vol. 15 Issue 2, p135, 14p
  5. What Events Are. By: Chemero, Anthony. Ecological Psychology, 2000, Vol. 12 Issue 1, p37, 6p
  6. Where Is the Information for Affordances? By: Gibson, Eleanor J.. Ecological Psychology, 2000, Vol. 12 Issue 1, p53, 4p
  • Learning
  1. Perceptual Learning in Development: Some Basic Concepts. Gibson, Eleanor J, Ecological Psychology, 2000, Vol: 12(4)
  2. Toward and Ecological Approach to Perceptual Learning and Development: Commentary on Michaels and Beek. By: Eppler, Marion A.; Adolph, Karen E.. Ecological Psychology, 1996, Vol. 8 Issue 4, p353, 3p
  3. Development of knowledge of visual tactual affordances of substance. Gibson, E. J., & Walker, A. Child Development, 1984, volume 55, 453 460.
  4. The World Is So Full of a Number of Things: On Specification and Perceptual Learning. By: Gibson, Eleanor J.. Ecological Psychology, 2003, Vol. 15 Issue 4, p283, 5p
  5. An Interview with Eleanor Gibson. By: Szokolszky, A. Ecological Psychology, 2003, Vol. 15 Issue 4, p271-281
  • Applications
    • Robotics
      1. Notes from Introduction to AI Robotics. By: Robin R. Murphy, (Section 3.4.3)
      2. Behavior-Grounded Representation of Tool Affordances. Stoytchev A. Proceedings of IEEE International Conference on Robotics and Automation (ICRA), 2005
      3. Toward Learning the Binding Affordances of Objects: A Behavior-Grounded Approach. Stoytchev A. Proceedings of AAAI Symposium on Developmental Robotics, 2005
      4. Automatic Object Recognition within an Office Environment. Wünstel M., and Moratz R. CRV 2004 Canadian Conference on Computer and Robot Vision, 2004
      5. Learning about objects through action -initial steps towards artificial cognition. Fitzpatrick P., Metta G.,Natale L., Rao A., and Sandini G. Proceedings of the 2003 IEEE International Conference on Robotics and Automation, ICRA, 2003
      6. Perception Driven Robot Locomotion. Lewis M. A. Journal Robot Society of Japan. 2002
      7. Contention Scheduling: A Viable Action-Selection Mechanism for Robotics? Andronache V., and Scheutz M. Proceedings of the Thirteenth Midwest AI and Cognitive Science Conference, 2002
      8. Exploring Unknown Structured Environments. Diaz, J., Stoytchev, A., and Arkin, R. Proceedings of the Fourteenth International Florida Artificial Intelligence Research Society Conference (FLAIRS-2001), 2001
      9. Development and Robotics. Metta G., Sandini G., Natale L.,and Panerai F. IEEE-RAS International Conference on Humanoid Robots, 2001
      10. Certain Principles of Biomorphic Robots. Lewis, M. A. and L. S. Simó. Autonomous Robots. 2001
      11. Further experiments in the evolution of minimally cognitive behavior: From perceiving affordances to selective attention. Slocum A.C., Downey D.C., and Beer R.D. From Animals to Animats 6: Proceedings of the Sixth International Conference on Simulation of Adaptive Behavior, 2000
      12. Responding to affordances: Learning and Projecting a Sensorimotor Mapping. MacDorman K. F. Proc. of 2000 IEEE Int. Conf. on Robotics and Automation, 2000
      13. Grounding symbols through sensorimotor integration. MacDorman K. F. Journal of the Robotics Society of Japan. 1999
      14. Case studies of applying Gibson's ecological approach to mobile robots. Murphy R. R. IEEE Transactions on Systems, Man, and Cybernetics. 1999
      15. Ecological Robotics. Duchon A. P., Warren W. H. , and Kaelbling L. P. Adaptive Behavior. 1998
      16. Ecological Perspectives on the New Artificial Intelligence. By: Effken, Judith A.; Shaw, Robert E.. Ecological Psychology, 1992, Vol. 4 Issue 4, p247, 24p
    • Other
    1. Affordance Theory for Improving the Rapid Generation, Composability, and Reusability of Synthetic Agents and Objects. Cornwell J., O'Brien K., Silverman B. G., and Toth J. 12th Conf on Behavior Representation in Modeling and Simulation, 2003
    2. An activity theory approach to affordance. Bærentsen K. B., and Trettvik J. Proceedings of the second Nordic conference on Human-computer interaction, 2002
    3. Artifact Intelligence: Yet Another Approach for Intelligent Robots. Takeda H., Terada K., and Kawamura T. Robot and Human Interactive Communication, 2002
    4. Human wayfinding in unfamiliar buildings: A simulation with cognizing agents. Raubal M. Cognitive Processing. 2001
    5. Objects, affordances ... action !!! Humphreys G. The Psychologist. 2001
    6. Capturing Temporality and Intentionality in Information Systems. Liu K., and Sun L. Proceedings of the Fifth International Workshop on the Language-Action Perspective on Communication Modelling (LAP 2000), 2000
    7. User interface affordances in a planning representation. Amant R. S. Human Computer Interaction. 1999
    8. Planning and User Interface Affordances. Amant Robert St. Intelligent User Interfaces, 1999
    9. A Formal Model of the Process of Wayfinding in Built Environments. Raubal M., and Worboys M. Spatial Information Theory - Cognitive and Computational Foundations of Geographic Information Science, International Conference COSIT, 1999
    10. Virtual intelligence from artificial reality: building stupid agents in smart environments. Doyle P. Working Notes of the 1999 AAAI Spring Symposium on Artificial Intelligence and Computer Games, 1999
    11. An Affordance-Based Model of Place in GIS. Jordan T., Raubal M., Gartrell B., and Egenhofer M. 8th Int. Symposium on Spatial Data Handling, SDH'98, 1998
    12. Affordances and the Practice of Industrial Design Engineering: Comments on Smets's Presentation Ad W. Smitsman A. W. Ecological Psychology. 1995
  • Experiments
  1. Events as Changes in the Layout of Affordances. Chemero A., Klein C., Cordeiro W. Ecological Psychology. 2003, volume 15, 1, pages 19-28.
  2. Perceiving walk-on-able slopes.Kinsella-Shaw, J. M., Shaw, B., & Turvey, M.T. (1992). Ecological Psychology, 4, 223-239.
  3. To Cross or Not to Cross: The Effect of Locomotion on Street-Crossing Behavior. Oudejans, R., Michaels, C., van Dort, B., & Frissen, E. 1996, Volume 8(3), pages 259-267.
  4. Perceiving affordances: Visual guidance of stair climbing. Warren, W.H. (1984). Journal of Experimental Psychology: Human Perception & Performance, 10, 683-703.
  5. Visual guidance of walking through apertures: body-scaled information for affordances. Warren WH Jr and Whang S. 1987. Journal of Experimental Psychology: Human Perception & Performance, 13 (3), pages 371-383.
  6. Eyeheight-Scaled Information About Affordances: A Study of Sitting and Stair Climbing. Mark L. S. 1987. Journal of Experimental Psychology: Human Perception & Performance, 13 (3), pages 361-370.<>
  7. <>Detection of the traversability of surfaces by crawling and walking infants. Gibson, E. J., Riccio, G., Schmuckler, M. A., Stoffregen, T. A., Rosenberg, D., Taromina, J. (1987). Journal of Experimental Psychology: Human Perception and Performance, 13,533-544.

What Is an Affordance?
Jones, Keith S.

  • Evolution of Gibson's affordance term
  • 2 categories:
    • Objects and events have no inherent meanings, everything created in animals mind.
    • Objects and events have inherent meanings, exploited by animal, without any mental calculation, direct perception view.
  • Evolution:
    • (1938) Valence term, motion perception, field of safe travel+minimum stopping zone. Connection with affordance, ratio of, Warren. Ratio of stair-riser height and leg length -> control of climbing behavior
    • (1947) Inherent meanings of optical variables, retinal motion pattern, optic flow
    • (1950) Possibility of perceiving spacial meanings
      • meanings and spacial properties not separable (color, texture etc.)
      • symbolic meaning separable and learned
      • meaning is inherent in something -> not created by perceiver
      • immediate awareness of the possibilities afforded by environmental objects.
    • (1966) Relation between perception and action
      • "When constant properties of constant objects are perceived, the observer can go on to detect their affordances"
    • (1979) The Ecological Approach to Visual Perception
      • Gives fundamental components of affordances
      • Environment offers affordances to animals, environment provides affordances to animals
      • Imply complementarity of animal and environment
      • Psychologists assume that objects are composed of quantities. What we perceive when we look at the objects are their affordances, not their quantities -> Special combination of quantities are not noticed.
      • Perceiving an objects properties and affordances is quite different.
      • Affordance: The affordance of anything is a specific combination of properties of its substance and its surfaces taken with respect to an animal.
  • Others:
    • Turvey (1992) formalized theory of affordances
    • Stoffregen (2000) contrasted affordances with events
  • At 2002, Ecological Psychology, North American Meeting:
    • Stoffregen critiques Turvey's definition and gives alternative
    • Claire Micheals discussed:
      • ontological status of affordances
      • their relations with actions
      • their relations with effectiveness
      • question of what level of description should be used
    • Chemero: Affordance is relation between ability of animals and features of the environment.

Affordances as Properties of the Animal-Environment System.
Stoffregen, Thomas A.

  • Affordances are emergent properties of the animal-environment system.
  • A critique of Turvey's formal definition of affordances (dispositional properties)
  • Properties of animal-environment system:
    • do not rely on dispositional properties (I think they rely on, but not completely)
    • should be consistent with direct perception.
  • Affodances are central to ecological approach to perception and action, lots of references.
  • Uncertainty in the exact meaning complicate attempts to use affordances.
  • In this article, a formal definition of affordances is proposed.
  • Definitions in prose:
    • What kind of things are afforded? Behaviors... ex:
      • Gap affords passage,
      • Stair with a certain propotion to leg length affords climbing
      • Rigid surface with respect to the weight of animal affords stance
      • Ball falls with respect to person's velocity affords catching
      • Time gap on traffic, safe crossing is afforded.
  • Definitions:
    • Affordances are relations between animal and environment with consequences in behavior
    • Affordances are properties of environment of an animal that have consequences for animals behavior.
  • Turvey: Affordance is property of environment, effectivity is property of animal -> Behaviors are actualized. However dispositional properties should be actualized whenever conditions are sufficient.
  • Formal Definition of Stoffregen:
    • Wpq = (Xp,Zq) ie person-climbing stairs system
      • p is property of X, q is property of Z
      • relation between p&q, p/q, defines a higher order property h.
      • h is said to be an affordance of Wpq iff
        • Wpq posses h
        • Neither X nor Z possesses h (exists only in animal-environment systems)
  • The dynamics of animal locomotion is influenced by the dynamics of surface support.
  • Not only leg length, but also strength, flexibility, skills...
  • Relations between the size and mass of the objects and the reach, grasp and displacement capabilities of the person combine not only whether prehension is possible but also how it should proceed. (compare this view with Michaels)
  • Definition of Behavior: Behavior is what happens at the conjunction of complementary affordances and intentions or goals. A given behavior will occur if and only if an affordance and its complementary intention co-occur at the same point in the space-time continium. From unlimited set of intentions, set of affordances determines those that can be satisfied at any given place and time. Architecture representation discussion! psychological choice function, m(h,i), this discussion is more central to behavior not affordances.
  • Perceiving that a given intention is not satisfied here, can motivate an animal to seek out conditions under which the intention can be satisfied. Motivates:
    • Exploration of the animal-environment system,
    • the acquisition of new perception-action skills,
    • prospective modification of the environment.
  • Implications on specification:
    • Identifying information that specifies affordances is a critical challenge. Gibson : "The central question..."
    • Describing the physical relations and showing that participants detect them does not explain how they are detected.. A lot discussion needed..

An Outline of a Theory of Affordances.
Chemero, Anthony.

  • Affordances are real and perceivable.
  • Affordances are not properties of either environment or animal but relation between environment and abilities of animals.
  • Meaning of perceptions?
    • Inferential theories: Meaning arise inside animal, based on interactions
    • Direct theories: Meaning is inside the environment. Not depend on meaning conferring inferences, simply gather information from a meaning-laden environment. If world contains not only physical appearance but also some more (in our case affordances), then it may include meanings.
  • Gibson: Affordance is a resource, environment offers any animal that has the capabilities to perceive and use it.
  • What kind of animal-relative properties?
    • Reed: The fundamental hypothesis of ecological psychology is that affordances and only relative availability of affordances create selection pressure on animals; hense behavior is regulated with respect to the affordances of the environment for a given animal. Resources in environment -> develop perceptual systems to perceive resources -> resources exist prior to animals. Tie evolution by natural selection and affordances closely.
    • Turvey: Affordances are dispositional properties of environment. Dispositions depend on possible actualizing circumstances, an object canbe edible only if there are animals that can eat and digest it. Affordances, in Turvey's preferred language, must be complemented by properties of animals. No selection pressure on animals. For example, body scale is the property of animals to which affordances of the environment related.
  • Summary:
    • Affordances are animal-relative properties of the environment.
    • Are these properties exist without animals?
      • Affordances are resources that guide through natural selection
      • Dispositional properties of environment that should be complemented by properties of animals.
    • Relevant properties of animals?
      • Abilities
      • Body scale
  • Chemero: Affordances are not properties! Affordances are relations between particular aspects of animals and particular aspects of situations.
  • Definition: Affords-Q (environment, organism), where Q is a behavior.
    • The relation affords-Q holds between environment and organism
    • The environment affords behavior Q to the organism.
  • Chemero: Perceiving affordances is placing features, seeing that the situation allows a certain activity. Environmental relata in affordances should be features not properties!!
  • Affords-Climbing (stepping ability, riser height) - Affords-Q(feature, ability)
  • Effectivity: Effectivities are defined as organismal complement to affordances qua properties of the environment. Dispositions are guaranteed to become manifest : soluble solid sugar will always dissolve in water in suitable conditions.
  • Turvey: effectivity <-> disposition
  • Ability: Functional properties of animals.
  • Chemero: affordance has connections with evolution, but not in a selectionalist view. Selectionalist view is inconsistent with animal-environment mutuality.
  • Niche: a set of affordances for a particular animal. An animals abilities imply an ecological niche. All abilities depend on basic orienting abilities.
Affordances: Four Points of Debate.
Michaels, Claire F.

  • 4 points:
    • ontological status of affordances , whether it exists without animals etc.
    • affordances are related to one others actions?
    • relation between affordances and effectivities
    • nesting of affordances, level of description
  • affordance: what environment offers to animals.
  • Affordances are not created in the act of perception.
    • Many scientists use term affordances to refer mental states regarding action possibilities
    • Affordances do not arise as a consequence of mental operations. They  are action-referential properties of the environment that may or not may be perceived.
  • Action relatedness?
    • Ground affords walking, stair-climbing, horizontal gap-step over, a nearby- object reaching
    • In Gibson, actions are absent
    • Cliff affords danger, air affords visual and auditory perception. etc..
  • Affordances is tied to intentions. dynamics of intention.
    • Visually guided action has two component problems. For example some situations affords swinging a bat but that the details of when and exactly how to swing are solved by organization that is assembled when on perceives the affordance.
  • Definition of Affordances:
    • Affordances are actions permitted an animal by environment objects, events, places...
    • Action: goal-directed movement
      • entails intention
      • detection of information
      • relation between information and control
    • Affordance is a multi-dimensional compound of properties.
    • Affordance exist independent of being perceived.
    • Perceiving an action -> some action can be engaged in himself.
What Events Are.
Chemero, Anthony.

  • The aim of the article is to oppose Stoffregen's argument on events such that we do not perceive events, thus there is no motivation for an event research in ecological psychology.
  • Events are defined as static and dynamic properties of objects and surfaces defined without reference to behavior and not scaled relative to action-relevant properties of animals. However, Chemero try to make some events action-relevant by defining the term as changes in the layout of affordances of the animal-environment system, in the examples of rising stair and widening gap, we perceive the critical point where affordance of the layout changes (we can climb/jump or not. This shows that we can perceive the events. In such a convention, events are perceived since they are action-related, personally scaled changes in the surroundings.
  • From my point of view, this is a direct consequence of a somewhat threshold based approach, where thresholds are directly known..
Where Is the Information for Affordances?
Gibson, Eleanor J.

  • Focuses on the significance of the perception of events on the perception of affordances in an ecological approach. Account of the development of the perception of events and affordances.
  • Not relevant to the discussion of information extraction from objects for affordances, but relevant to discussions on events..
  • Members of air force, importance of being in high speed plane
  • Gibson: Information for perceiving the layout in motion and events occuring over time, Gibson interested in the case of a moving observer finding a safe path and avoiding obstacles.
  • Information for an affordance is to be found in events that include the relevant environmental features, the activity of the organism, and the consequences that ensue as well as the relations among these.
  • James Gibson in 1966, putting them into categories and subcategories, noticing their similarities and dissimilarities.
  • Minimal exploration behavior of babies, such as sucking and looking, that enable them to their education. Learn their own abilities to control events, and outcome... Eppler, 1993

Perceptual Learning in Development: Some Basic Concepts.
Gibson, Eleanor J
  • This paper is a rare one, in which learning of affordances are discussed. Eleanor Gibson based her perception learning theory on differentiation, instead of a process called association. She defines affordance learning as the learning of relation between "organism's power of control" and "some offering of the environment". Learning is a process of selection, not construction from smaller pieces.
  • She asserts that this relation is learnt during development, and gives examples from infant studies.
  • In her own words, perceptual learning entails discovering "distinctive features" and "invariant properties of things and events".
  • She underlines that even new born infants are ready for active probing for i)gathering external information, and ii)exploring body's own capabilities. James Gibson locates information in ambient array of energy, and Eleanor Gibson says "this array must be searched by the appropriate perceptual systems to extract invariants". Exploratory activity is crucial in her theory, as an example, infants selects novel information instead of old/repeated experience. By the means of this activity, organisms gradually.
  • Perceptual learning is a selective process, and selection process rests on two principles:
    1. for an Affordance Fit: Fit between ctions performed by an infant and the consequences at the end. Learning to grasp, and carrying to the mouth for further exploration. In order to grasp, first infant should learn if it is reachable, therefore, learn its own arm length, the optimal or maximal size for grasping etc. Additionally, in order to reach an object, it should learn the affordance of locomotion.
    2. for Unity, Order, and Economy: "Economy of action and reduction of perceptual information stand out as principles of selection for increasing specificity." .. " Perceiving an object as a unit is a case of detecting order and reducing information..." Talks about intermodal redundancy.
Toward and Ecological Approach to Perceptual Learning and Development: Commentary on Michaels and Beek. By: Eppler, Marion A.; Adolph, Karen E.
  • list of papers on ecological psychology, perception learning, and perception especially in infants:
Development of knowledge of visual tactual affordances of substance.
Gibson. E. J., & Walker

  • Experiments on infants for study of affordance of rigid and elastic objects, and establish connection between haptic and visual systems.
  • In first experiments, infants are given identical rigid and elastic objects to play with, in a dark room. After the exploration phase, objects are taken and motion pictures of not identical but similar rigid and elastic objects. Infants pay more attention to the objects that they did not see but handle before.
  • In experiment 2, the objects are given in a lighted room. The results are similar to the first experiment. The question is that in first experiment, is it experience or direct detection of intermodal invariants that drive affordance?
  • Experiment 3 was performed on very young, 1 month infants. The objects was concealed by hand while babies mouthing the objects. It is seen that babies pay attention to the novel objects.
  • In all experiments, there is a tendency to look either novel or handled objects. This shows the existence of  the detection of intermodal invariance on objects.
  • Infants while handling, handle elastic objects and strike rigid objects to ground, and while mouthing chew elastic objects. However, no real discussion is performed on affordances, detection of affordances, learning of affordances...

The World Is So Full of a Number of Things: On Specification and Perceptual Learning.
Gibson, Eleanor J.

  • Main emphasis is on the discovery of information that specifies affordances, in perceptual learning. The aim of the paper is to discuss the importance of perceptual learning: The process of discovering information that specifies an affordance is perceptual learning
  • Differentiation is the key term.
  • Perceptual learning is not properly defined as association or as and addition of any kind, as a response to a stilmulus, or as a "representation" of anything. An active perceiver has the tasks of extracting the information that specifies relevant events and especially, of detecting the information that specifies an affordance of the environment relevant to the perceiver's species, needs, and powers. Learning to detect the information that specifies such a relationship is perceptual learning. Narrowing down from vast manifold information to minimal, optimal information for the affordance of an event or object or layout.
  • Infant learning and infant labratory Discover invariants of things and places.
    • Learning about communication, learning about objects, and learning about locomotion. (most exciting comes with the advent of locomotion)
  • Perception learning is not a matter of enriching the input but of differentiating the information.
  • Reference: An Ecological Approach to Perceptual Learning and Development.
  • The baby (after learning) learns how to use its perceptual systems for obtaining information about what is going on nearby.
  • Exploratory systems that are available very early are mouthing, listening and looking. Exploratory strategies are greatly extended by increasing abilities to reach, grasp and engage.
An Interview with Eleanor Gibson.
  • Learning development of affordances.
  • Not only static perception.
  • Experiments with baby animals.
  • Only Bob Shaw does research in cognition in the context of ecological psychology.
  • Visual Cliff Experiments
Notes from Introduction to AI Robotics.
Robin R. Murphy (Section 3.4.3)

  • Affordance may be viewed as a percept that guides behavior. Presence of red to a hungry baby arctic tern releases the feeding behavior. Red: guide as well as release an action.
    • High object and ramp? object height guides action?
  • poppy color evolution for bee bandwidth (evolution of environment-agent system...?)
  • Why attractive from robotics point of view?
    • Affordance is directly viewable, not require memory, inference mechanisms...
    • Ball thrown, optic flow-> simultaneous ducking behavior
    • Time to contact
  • Not form or structure BUT function... Stark and Bawyer
    • GRUFF identifies chairs by function rather than form, chair is ... which serves functionality of sittability, no memory, no inference, no interpretation?
    • seat size, height etc. should be memorized
  • Neisser: 2 perceptual systems
    • Direct perception
    • Recognition
    • When solving a problem, it should be determined if affordances are sufficient or not for the problem. If not, one should 'certainly' employ more sophisticated methods.
Exploring Unknown Structured Environments.
Diaz, J., Stoytchev, A., and Arkin, R.

  • In this article, affordance is not applied explicitly, and mentioned very abstractly. Perceptual algorithms may be related to affordance discussions, but they are very general and straightforward techniques.
  • The aim of the paper is the exploration of structured environments, whose inherent structures are statically embedded to the control of the robot, which is a severe restriction for studies whose aim is to develop systems for search, rescue missions.
  • Literature on docking motor schema, which seems nice. Arkin and Murphy (90), according to some endpoints, construct vector fields, where robot make smooth moves.
  • In corridor detection, based on laser range sensor readings, hallway angle and width is calculated in order to navigate robot in the middle line of the corridor. In door detection, again absolute values, end points of door openings and width is calculated. Box filters with robot-based length and width is used for detection of appropriate openings.
  • Usage of laser range scanner in perception of corridor and doors may be used...
  • Doorway endpoints repels robot and hallway attracts, a potential field is created... POTENTIAL FIELDS!
Case studies of applying Gibson's ecological approach to mobile robots.
Murphy R. R.

  • Main emphasis is on the application of affordances in robotics as direct perception, where is can be alternative to model-based approaches. It says it is an art rather than science, as noted in all reactive implementations, and affordances are compatible with reactive robotics.
  • Literature on use of optic flow in robotics is represented.
  • Literature on functionality of objects in the context of affordances is represented.
  • Neisser is important since depending on this neurological and cognitive studies, he suggests two perceptual systems, for direct perception and for recognition, where second one accounts for semantic content (ie. your coffee cup). This in turn suggests that affordances are not applicable for all behaviors.
  • Murphy developed four step methodology for identifying affordances:
    1. Define the task in order to establish whether the behavior is a suitable candidate for affordance-based perception.
    2. Define the ecological niche for the robot.
    3. Isolate potential affordances either by examining biological processes or by extensive consideration of the perceptual context.
    4. Implement the perceptual process for each affordance and test over the range of operating conditions.
  • Three case studies: 
    • Fine positioning: The problem is to dock with a workstation. First robot uses coarse ballistic movement to get in the range, then use affordances for fine-positioning. The inverse perspective tranform, which is based on distortions in size and shape of landmarks might be used, but it is very sensitive to segmentation errors. Inspired from fish schooling behavior and employment of number of activated cones that plays important role in fish perception, following method used: Surface with a uniform texture appears differently from different orientations. Therefore, the region is divided into two, and the ratio between two halves give directions.
    • Path Following: A winner white line path following algorithm is given. The algorithm is solely based on majority of the brightness pixels. Sonars are used to differentiate white objects from white line. In the discussion, direct perception is good for predictable environments, and affordances should be carefully designed, that the system should not be brittle. (Ecological niche)
    • Pick up the Trash: There are red, white cans, and trashes that should be put in. Colors serves as affordances. There is no need to model because colors are distinctive, and there was no reason to label cans, only approach to the closest can. However, system is again brittle, in different lighting conditions, brown surrounding boxes are similar to red cans, and attract robot.
  • In discussion, the information on color of cans or color of line or textures are innate and no requirement for explicit perceptual models.
  • My result is the article is on direct perception but could hardly be used in affordance based robotics. But it gives clues on the difficulties of designing of affordances, importance of complexity of tasks, Neisser's division of perception system, and the importance of ecological niche.
Ecological Robotics.
Duchon A. P., Warren W. H.

  • This article mainly emphasizes the striking parallels between ecological psychology and new trends in robotics in terms of interaction with environment. Mainly optical flow is employed in  obstacle avoidance and game of tag on two different robots. (Louie & Ramona)
  • Robotic research mostly employs sensors other than vision because vision problems are usually ill-posed. However, they allege, instead, by active vision, these problems would become well-posed. He suggest vision as the most promising sense relating robotics and ecological psychology. Following Ballard's (1991) question, 'what is vision for? ' they concluded animals may not need to model their environment in order to perceive the relationship for survival.
  • direct perception: 'the animal has direct knowledge of, and a relationship to, its environment as a result of natural laws. Direct perception is usually applied to motor behaviors, BUT animals not respond stimuli, choice among many possibilities is done.
  • They gave a introductory description of optic flow and motion control laws, and present related work on control rules, like control of balance, steering, baking in humans, and control of flight in flies. Control laws are mostly referred to Warren. Authors additionally give emphasize on the importance of action mode, when governing different control laws, and relate this to affordances.
  • Authors define the term ecological robotics as the field which applies principles of ecological psychology to mobile robot design. They gave some principles of ecological robotics:
    • Animal-environment system should be treated together
    • Dynamics of the system has main role on animal's behavior.
    • Task is to map available information to control parameters in order to achieve a desired state.
    • "The environment provides enough information to make adaptive behavior possible"
    • No central model is need, but memory for specific tasks may be employed.
  • Detailed formulas of control laws on obstacle avoidance, wandering, fixating the target, escaping, and chasing are given, and described. Some triggering mechanisms for activation of these behaviors are determined according to optic flow. Additionally segmentation from optic flow is used for escaping and chasing. Results showed that there are some drawbacks in usage of segmentation, and it requires careful design.
  • Centering response: bees fly to equate optic flow in the lateral positions of each eye.
  • Balance strategy is used in obstacle avoidance from the motivation, closer objects give rise to faster motion across retina than farther objects. It is found to be resistance to noise. However it is inept when robot is encountered with corners since it has same properties with hallway.
  • 2 reflexes for emergency : 1- intensity reflex (when no optic flow available) 2- eta-reflex (when too close to an object).
  • If FOE (focus of expansion) is false due to imperfect alignment of camera with wheels, is could be adapted.
  • Some literature on robotic implementations of similar ideas are given, and an additional one for insect literature is given.
  • The behavior when robot passing through apertures, is obtained and discussed in terms of optic flow and speed of the robot.
  • Some literature on control laws:
    • Subsumption architecture
    • Simultaneous acting of controls with varying thresholds (Virsik 76)
    • Simultaneous acting with time-scales (Wagner 86)
    • dynamics (Schoner 1995)
    • schemas (Arkin 1990)
  • When Wanderer moves at an angle to Watcher's heading, because of the resulting flows, target is misinterpreted to expand and thought it is close. One possible explanation for prey's zigzag movement.
  • Literature and references on action-perception robotic research and planning.
    • Aloimos, Brooks, Coombs, Horswill, Pfeifer,and Sandini
    • Mataric and Meeden
  • One conclusion or motivation 'know that' first requires one to 'know how'. First avoid, then know what is it..? BUT both are required.
Ecological Perspectives on the New Artificial Intelligence.
Effken, Judith A.; Shaw, Robert E.

  • In this article authors make reviews of two books, which study robotic problems in ecological perspectives. Both Connell's and Beer's books criticizes traditional AI approach, and opposes of its deliberative reasoning approach, symbolic representations and restricted application domains. Beer says domain-specific intelligence is often totally inept outside that domain, and could not form a basis for intelligent behavior.
  • They oppose the view of deliberative reasoning and instead they propose adaptive behavior, which implies animal's short term survival. From traditional AI view, internal models should be constructed as world maps, however Connell suggests an environmental consequence of an action taken by the robot is perceived and used to modify and coordinate behavior directly without intervening mental models or plans.
  • In explaining reciprocal fine-tuning relation of detection and control (action), authors compare European and Trukese navigators in their exploration patterns.
  • As a last word Beer in his book explores encoding intelligence as adaptive behavior, and Connell demonstrates goal-based robot "behavior without representation".
Review: Intelligence as Adaptive Behavior: An Experiment in Computational Neuroethology. Randall D. Beer.
  • He designed a cockroach who is mainly governed by its perceptual sensitivity, by mimicking reflexes, taxes, fixed-action patterns, and motivational control.
  • Beer is influenced from
    • Maturana and Varela's (1980) structural congruence theory which is about animal-environment relationship.
    • Schoner and Kelso's (1988) dynamic patterns
    • Brook's ideas (1986)
    • Braitenberg's synthetic approach, which dictates the law of uphill analysis and downhill invention. Synthetic approach is reminiscent of Grey Walter's Elmer and Elsie electronic tortoises (1940), which seeks light, charge, explore, become hungry, and seek light again.
  • One of the main emphasis is on the modeling of the simple whole animal in their environment, instead of modeling a subsystem of a complicated animal in a restricted domain. This would give better insight on natural systems. Beer defines this approach as computational neuroethology.
  • Shaw and Bransford (1977) evolution tunes perception systems to be most sensitive to the invariants (is one of the greatest adaptive significance to the animal).
  • Control is established through a number of mechanisms (neurons)
    • A central pattern generator for rhythmic movements of leg
    • Stance and swing motor neurons for swinging when eating food
    • 6 individual leg controllers for avoid walking when moving
  • Control of the gaits cannot be designed statically, but should be made dynamically, heavily depending on the stimuli (exproriospecific information)
  • There is no mechanism for compensation of a lost leg, Effken et. al. suggests a mechanism where reflex detectors perceive absence of contacts, and make use of another gait control. Beer suggests redundant structures that could be recruited in case of emergency.
  • Effken et. al. suggests directions for future work on robotics, where he gave termite nest building mechanism.

Review: Minimalist Mobile Robotics: A colony-Style Architecture for an Artificial Creature. Jonathan H. Connell.

  • Problem in traditional robotics stems from the centralized nature of world models and path planners since all sensory inputs should be condensed for the model. Connell relies heavily on vertical decomposition of tasks and distributed perception and control of independent agents for tasks, which may in turn limit the overall performance.
  • Control is solely depend on local information, no mental or world models involved. This approach may be enriched with incorporation of global intentional constraints for future development. Independent agents work with local info but these agents should monitor the plan along the way.
  • Use of world as its own representation, organism-environment duality.
  • There is no overall plan and representation, but embedded partial representations and sketching plans.
  • Decend, Bounce, Surface and Over behaviors for control of hand to pick-up can above a table. The robot's plan is actually derived from the sequence in which the environment allows behaviors to be triggered:
    • raise hand, extend hand, hand down, search for can, when approach -> local proximity sensor allignment
  • The trajectory is guided by environmental constraint, very flexible in case of changes.
  • Six independent behaviors are included into a sort of subsumption architecture. Insert new levels.. With such a mechanism, complexity of behavior of arm controller reduced.
  • A Gibsonian view on object retrieval. When robot perceive a can, it approaches to the can. If can's image is greater than a predefined patch, robot loses interest to the object.
  • Direct coupling of perception-action, robot concentrates only on that part of the environment directly relevant to current activities.
  • For future:
    • Local minima problem can be overcome by acting on no showing of progress.
    • Detection of invariants (exploitation of Gibson's affordance concept). Structural invariants are those properties that remain unchanged under transformation. Gibson: When the constant properties of constant objects are perceived (the shape, size, color, texture, composition, motion, animation, and position relative to other objects), the observer can go on to detect their affordances. The design trade-off: determining invariant structure that specifies affordance and building robot with the capability to detect the invariants in its world. Critical preliminary for learning. Which affordance requires which invariant?
    • If usual path is blocked, develop alternatives
    • No adaptation... An alternative: priority setting procedure, theta-space summation alternative where behaviors are rated in all possible directions.
  • Effker and Shaw emphasizes the importance of dual work performed by ecological psychologists and robotic researchers.

Human wayfinding in unfamiliar buildings: A simulation with cognizing agents.
Raubal M.

  • Afffordances : 1- pyhsical, 2- social-intitutional, 3- mental affordances...
  • interested in immediate sensing of information presented
  • no ability to learn, everything is explicitly given
  • transition graph and commonsense reasoning on decision points
  • extracting meaning of features directly
  • "Gibson describes the process of perception as the extraction of invariants from the stimulus flux. Surfaces absorb or reflect light and Gibson's radical hypothesis is that the composition and layout of surfaces constitute what they afford"
  • Gibson is criticized by neglecting process of cognition.
  • Gaver "a person's culture and experience and intentions also determine her perception of affordances".
  • path's affordance: 'go-to' affordance, sign: 'reflect light, match', decision point: 'search'

Virtual intelligence from artificial reality: building stupid agents in smart environments.
Doyle P.

  • This article is devoted to the development of intelligent agents in virtual worlds, especially for online games. The aim is to increase the believability of the agents, giving adaptation abilities to agents in changing worlds, equipping agent with learning abilities, enabling agent to direct perception and means of usage of environmental features, and enabling agents to have completely different characteristics when encountered with different situations.
  • Affordance, is only used as an abstract concept, author described affordances and nice features that it provide. The world is its own best model , from Brooks, construct a basis for the ideas in the paper. For example agents would learn from  experiences through direct queries rather than pure reasoning, but this point of view has loose relation with Gibsonian perception approach.
  • Initially, the characters only know the world's basic dynamics, by using the annotations given, they will become believable and living characters. They will naturally evolve as unique individuals.
  • Additionally updating [world] annotations is easier than reprogramming each agent.

Events as Changes in the Layout of Affordances.
Chemero A., Klein C., Cordeiro W.

  • The motivation is to show that event as changes in the layout of affordances are perceived.  The question is on the value of event perception research.
  • Change in the layout of affordances: Appearance and disappearance of affordances
  • 2 types of events:
    • Ecological events changes in action-neural physical magnitudes
    • Physical events static or dynamic properties of objects.
  • Judge the critical point where gap not affords passing over (gap crossing affordances disappear)
  • 3 main experiments:
    • step-across judgment task with a platform moving away from them
    • step-across judgment task with static platforms
    • and actual stepping task to measure stepping length.
  • Results show that there is a high correlation between critical points of first two experiments, and real stepping length.
  • The discussion is on the notion of event, should changes in the layout of affordances regarded as events or not.
Perceiving walk-on-able slopes.
Kinsella-Shaw, J. M., Shaw, B., & Turvey, M.T.

  • The main emphasis of the paper is on the distinction of visual and geographical slant study. The previous slant perception has limitations, later one has implications on affordance study since it underscores the importance of action-relevant measures.
  • In geographical slant, the focus is on a surface's inclination relative to the surface of the earth where in optical slant, absolute measures are considered, ie with respect to another plane.
  • In the first experiment, the slope of the surfaces are incremented and decremented by a participant in order to determine the critical walk-on-able point. It is observed that independent of the distance of the participant, he can judge the critical point correctly.
  • In the second experiment, the relation between haptic and visual contacts with slanted surfaces are compared, participants were able to judge the correct visual slant, with respect to a haptic slant applied to one of his foot.
  • My own notes: The application to the robotics would be to learn the critical slant of robot, and than robot's weight may be changed (robot may carry a heavy object), and affordance critical slant should be adjusted accordingly.
To Cross or Not to Cross: The Effect of Locomotion on Street-Crossing Behavior. Oudejans, R., Michaels, C., van Dort, B., & Frissen, E.

  • Main emphasis of the paper is on the comparison of crossability affordance of a street between in a stationary state or pre-walking state.
  • For some affordances, body scaling is not sufficient, and scaling of environment to action capabilities is required.
  • In the end of the experiments (observations), it is seen that moving people judge the critical value of crossing much better than in stationary because affordance entails a relation between time gap and walking speed, and perception of the walking speed is clearly not accurately available in stationary state.
  • Gibson : We must perceive in order to move, but we must also move in order to perceive. Clearly while walking we perceive our ability about walking. Additionally while walking we possess enhanced perception of distance and enhanced perception of time to arrival because of the optic flow.

Perceiving affordances: Visual guidance of stair climbing. Warren, W.H.

  • Main emphasis of the paper is the affordance of objects or features, where constant critical points (phase transition to a new action) and optimal points are perceived, and these points are body-scaled metrics with respect to environment and experiments are performed for stair climbing case. Optimal points are searched in terms of minimum energy consumption.
  • The motivation is to understand how animals and humans could efficiently visually guide in a cluttered environment.
  • In a work of Gibson (1983), infants exploration of traversable and non-traversable surfaces.
  • A path affords locomotion?
    • surface rigidity in relation to body mass,
    • surface contours in relation to gait,
    • passage width in relation to body dimensions, and so on.
  • Perception of relationship between environmental properties and properties of its own action system: Intrinsic or 'body scaled' metric for analysis of visual information.
  • Affordance studies for animals:
    • Frog  detects apertures that afford passage for its specific size.
    • Praying mantis attacks according to prey radius and it's forelimb.
    • East-coast limpets: the affordance of attack and retreat are determined by relative size of predator and prey.
  • Affordances do not depend on solely geometric variables but also kinetic and dynamic variables such as mass, force, friction, elasticity.
  • In stair climbing case: the functionally relevant variables are: step frequency, riser height, stair diagonal, leg length, and body weight.
  • The intrinsic unit, the dimensionless ratio, pi is defined to be the ratio of environment related properties and animal related properties. Through experiments this pi ratio is taken to be R/L where R is riser height and L is leg length. If this ratio is too high, it is very difficult to climb the stair; if too low,  the time and energy will be large in order to reach a certain height.
  • 3 experiments are performed:
  • In first experiment, visual perception of critical riser height is first estimated in pi ratio, then experimented, and at last observers are asked to judge whether each stairway looked climbable or not. Both short and tall observers judge critical height, in a constant pi.
  • In second experiment, energetic of optimal riser height, the height of the riser is tuned, and participants are asked to walk on the stairs. According to the total oxygen consumption, a pi value is again computed. This constant is found to be same for tall and short participants.
  • In third experiment, visual perception of optimal riser height is studied, and visual preferences of tall and short observers are found to be compatible with the pi number of second experiment.
  • The intrinsic optical information is underlined in the discussion part, since it is available in natural viewing conditions.
Visual guidance of walking through apertures: body-scaled information for affordances. Warren WH Jr and Whang S.

  • The previous work, which was devoted to understand the perception of constant optimal and critical points on climbability affordances, is extended to the analysis to visual guidance of walking through apertures. Passability judgments are studied under static and moving viewing conditions. Aperture-to-shoulder width ratio is found to be the body-scaled constant which results in shift in behavior (transition from frontal walking to body rotation.)
  • Affordance problem: visually controlling body rotation when walking through narrow apertures.
  • Evidence for body-scaled perception: Hallford (1984) Observers are more accurate and had higher confidence when judging whether they could grasp rectangular objects with one hand than when judging their widths relative to a standard object that was first viewed and then held during the judgments.
  • Other works:
    • Optimal and maximal sitting height (Mark and Volene)
    • Maximal reaching distance (Carello, 1986)
    • Infant climbing stairs (Ulrich, Thelen, and Niles, 1986)
    • Infant traverability of surfaces (gibson, Riccio, Stoffregern..)
  • Experiment 1: Critical Aperture Width is analyzed for rotation of body, done with narrow and broad shoulders. Aperture width and group are significant, and there was also a main effect of speed. The result is intrinsic scaling eliminates group differences. and critical point is a constant for the ecosystem, regardless of scale changes in the size of the actor.
  • Experiment 2: Visual Perception of Critical Aperture Width is analyzed. Two different conditions, first one is static condition with monocularly preferred eye, and second condition is moving condition with binocular view. In both conditions, for varying aperture sizes, participants are asked whether the gap is passable or not. In this experiment group and aperture width are found to have a main effect but viewing condition effect is found to be insignificant. The aperture-shoulder width ratio is found to be a bit smaller than first experiment, most probably because of the setting of the experiments.
  • One main results is that, despite the fact that more visual information is available in the moving than in the static condition, there was no main effect, and perceptual category boundary values are nearly identical. Experiment 3 are conducted in order to understand this unexpected result, is prior exposure the main effect?
  • Experiment 3: is on the available information (perception) nature, is it intrinsic or extrinsic? In the first paragraphs of this experiment, the importance of eyeheight information and what clues it may provide to us about height/width/distance of surrounding objects is discussed. After this discussion, in experiment environment, the eyeheight of participants are changed and this modification with its known observable results is conceived from participants. The aim is to try to change intrinsic information (eyeheight ratio) without changing absolute values (absolute width of apertures). Since most probably people use this intrinsic information (eyeheight ratio) when perceiving the width of the aperture, their estimation of the size of aperture in terms of passability affordance, is altered; observers overestimates the size of the apertures in the raised floor condition.
  • Although observer make false judgments on width of the apertures, distance values remain invariant independent of the eyeheight ratio.
  • There are some work on animals:
    • Ingle and Cook (1977) found similar cues on frogs jumping through apertures.
    • Lock and Collett (1980) found similar cues on toad move through apertures.
  • The general results are:
    • Static monocular vision seems sufficient for judgments of passability.
    • Animals use intrinsic information (perceive aperture width as ratio of eyeheight).
Eyeheight-Scaled Information About Affordances: A Study of Sitting and Stair Climbing. Mark L. S.

  • The main emphasis is on the intrinsic or body-scaled information (scaling with reference to eye-height).
  • The extrinsic traditional approach says that for example distance is scaled in meters, and we make a re-scaling according to our demands and body-size. However, in intrinsic view, there is no such metric absolute information that we obtain, but instead we perceive them either far or close to reach.
  • Climbing on case, Warren find constant proportion. 
    • Sitting Mark and Vogele constant wrt. person's leg length. 
    • Reaching  Solomon et. al.
    • Grasping Hallford.
    • Passing through aperture, Warren and Wang.
    • Children's climbing Ulrich, Thalen and Niles.
    • Style of contact Lee.
    • Catching behaviors Micheals, Prindle and Turvey.
  • Experiment 1: Observers utilize eye-height scaled information about surface height in determining whether surface can be sat on and climbed on. For each subject, judgments of perceived eyeheight, in terms of max sit and climb cases, decreased monotonically with viewing distance. In contrast to these, estimates of object size did not decline as a function of viewing distance. Changes in judgment of two action boundaries are closely correlated with perceived eye-height.
  • The motivation between the relation of eye-height and viewing distance is a suggestion that observers tend to underestimate their eyeheight as a function of their distance from surface at which they are looking???
  • Experiment 2-3: A block is wore by observers, therefore, they were aware of the change in their body-abilities and eye-height dimensions. Results show that observers use intrinsic eyeheight-scaled information in determining..

Detection of the traversability of surfaces by crawling and walking infants. Gibson, E. J., Riccio, G., Schmuckler, M. A., Stoffregen, T. A., Rosenberg, D., Taromina, J.

  • The main objective of the experiments is to study affordance of surfaces, via rigid and deforming surfaces. The experiments are performed on both crawling and walking infants, and effect of visual and haptic information is tried to be understood.
  • In order to perceive the affordance of sth, we should also detect the constraints imposed by own statures (ie. Warren's stair height, observer's height experiment).
  • The importance of visual and haptic information, and any dominance of visual or haptic information is looked for.
  • The rigidity of surface is defined as its resistance to deformation.
  • Experiment 1: Two textured surfaces, deforming one is agitated by an unseen experimenter, giving a wavy appearance. In the results section, rigid surface tended to have a shorter tendency for initiation of traversal for walkers. There is no differentiation between surface latencies for crawlers. The next experiment used dual way, and infants are asked to select rigid or deformable surfaces. Again the behavior and latency of walkers and crawlers differentiated.
  • Other 3 experiments are performed in order to understand the importance of visual and haptic information, by means of contradicting the optical information for deformability, giving surfaces which does not give any visual clue on its deformability.
  • The details of the experiments are not relevant to affordance based robotic research for the time, since there are very detailed discussions on other factors like age, effect of novelty, etc.
  • BUT on conclusion for is may be agents employ both haptic and visual information when deciding the affordance of traversability of surfaces.

eXTReMe Tracker