Contributed by: Xabier Barandiaran, University of the Basque Country

Overview
The term autonomy (from the greek auto = self and nomos = law) refers to the capacity of living and cognitive systems to generate their own laws or norms. The field of Autonomous Robotics has used the term to denote that a mobile robot is capable of operating without external human control and it is usually associated with situated approaches to cognition. However a full sense of autonomy requires more than non-human remote control: an autonomous agent is capable of generating its own regulatory principles. Autonomy has been proposed as a principle of biological organization: as a class of mechanistic organization characteristic of living and cognitive organism. Closelly related to the notion and study of self-organization and cognition a set of important consequences unfold from the autonomous approach for cognitive science: it puts the organism's "point of view" at the center, its biological embodiment and the recursive and distributed dynamics that it generates.

Introduction
When we look at organisms in nature, we observe their behaviour, their capacity to adapt and survive and we attribute some agency to the organism, as if the cause of behaviour was somehow coming from within the organism. In fact the generation of adaptive behaviour has a great asymmetry between internal and external (environmental) factors, so that although influenced by external conditions behaviour appears driven from inside the organism. But if we "open" it searching for the origin of such behaviour we wont find any central processor or explicit program specifying it (no will, homunculi, or locus of intentional cause), but, on the contrary, we find highly recursive and integrated mechanisms such as metabolic or neural networks in a constant ongoing activity. We say that organisms (as a whole) are autonomous or have a certain degree of autonomy on the generation of adaptive behaviour (and the construction and preservation of their own biological structure).

All organisms appear endowed with the capacity to continuously generate and regenerate their component processes, to set their own goals without external control and to adaptively and selectively modify their internal and interactive processes accordingly. It is their autonomy that strikes as a characteristic yet elusive property: i.e. their capacity to create their own norms, constraints and regulation in a recurrent way, creatively playing with the laws of physics and chemistry against thermodynamic inertia, while continuously reproducing their own identity. Since early Aristotelian conceptualization of nature, organisms have been distinguished by the holistic dependence between matter, form and function in them; by their unity as tightly integrated systems, possibly with an intrinsic teleology, a continuous regeneration of its ''raison d'être'', of its causes. Kant was among the firsts to recognize the challenge that the understanding of such systems supposed for the human mechanistic mind set-up (specially after its shaping by Cartesian rational materialism and Newtonian corpuscular physicalism). Yet Spinoza built good part of his philosophy of nature on top of the notion of conatus: organisms’ intrinsic drive towards self-assertion and self-maintenance. While early modern biologists (Shawn, Goethe and the early naturalists) focused on the cell as a unitary whole endowed with fascinating properties (far beyond what the mechanics of the time could achieve) the fast growing science of chemistry triggered a long standing debate between vitalists and reductionists and the mainstream scientific community was soon laden to the fast-growing achievements of reductionist-atomistic methodologies.

It is not until the development of some early-cyberneticist ideas during the 70s that the concept of autonomy (closely related to the notion of network, self-organization and closure) came to be reformulated so as to be introduced back again into the scientific discourse. Today, the concept of autonomy (striped away from its vitalistic connotations) stands, for many, at the center of the sciences of complexity, systems biology, robotics and artificial life. But autonomy as a property or condition of the organism is not easy to specify and make explicit. The term autonomy in relation to biology, adaptive behaviour and cognition has been used in many different ways ranging from a foundational notion for biological organization (Varela, 1979) to a label for a set of engineering constraints in robotics (Maes, 1991), but there is still no clear cuantitative definition of autonomy so that it can be directly applied to a dynamical modelling and simulation of adaptive behaviour. In addition the relation between autonomy, adaptation, its biological grounding and its robotic application is often conflictive and misleading in the literature.

Autonomy across the Literature. A Quick Overview
The term autonomous robotics has been used since de 90s (Maes, 1991) to refer to a set of engineering constraints on the construction and testing of robots, thus labeling a style of robotic research in cognitive science and engineering. Such constraints include conditions like no remote control of the agent, no external energy supply, motility in the robot, no human intervention in robot task solving or real-time response in real-world environments. Close to situated robotics (Brooks, 1991), autonomous robotics highlights physicality, embodiment, situatedness and dynamicism versus abstract, virtual and formal approaches to robotics and artificial intelligence in which agents operate in controlled formal or virtual environments or in toy like worlds without dynamical constraints. As a consequence of its embodiment and situatedness in real-world environments when talking of autonomy the emphasis is often put on the viability of the robot as a task achieving agent: a self-generated and robust capacity to respond to environmental changes.

The practice of autonomous robotics has forced some engineers to go beyond the specification of a list of engineering constraints and to develop a more elaborated notion of autonomy that specifies the kind of interaction process that is stablished between the robot and its physical environment, the dynamic structure and properties of the controll mechanisms and the underlaying consequences for cognitive science and epistemology. That is the case of engineers like Tim Smithers (1997,1995) who has developed a theory of behavioural autonomy in robots challenging information processing approaches to cognition, Randall Beer (1997,1995) challenging representational approaches to cognition or Eric Prem (2000,1997) who has put the emphasys on epistemic autonomy "the system's own ability to decide upon the validity of measurements" (Prem, 1997) a process that cannot be reduced to formal aspects, given the physicality of the measuring process, its pre-formal nature.

This authors have been greatly influenced by the biologist Francisco Varela whose definition of autonomy goes as follows: "Autonomous systems are mechanistic (dynamic) systems defined as a unity by their organization. We shall say that autonomous systems are organizationally closed. That is, their organization is characterized by processes shuch that (1) the processes are related as a network, so that they recursively depend on each other in the generation and realization of the processes themselves, and (2) they constitute the system as a unity recognizable in the space (domain) in which the processes exist." (Varela, 1979, p.55). Autonomy is thus defined as an abstract systemic kind of organization (relation between components and processes in a system), a kind of self-maintained, self-reinforced and self-regulated system dynamics resulting from a highly recursive network of processes that generates and maintains internal invariants in the face of internal and external perturbations, a process that separates itself from the environment, defining its identity; i.e. its unity as a system distinguishable from the surrounding processes. This abstract notion of autonomy is realized at different biological scales and domains defining different levels. As a paradigmatic example "life" is defined as a special kind of autonomy: autopoiesis or autonomy in the physical space. In turn "adaptive behaviour" is the result of a higher level of autonomy: that of the nervous system, producing invariant patterns of sensorimotor correlations and defining the behaving organism as a mobile unit in space.

But Varela's perspective on autonomy (although highly influential) has been recently critised by its emphasis on closure and the secondary role that system-environment interactions play in the definition and constitution of autonomous systems. Introducing ideas from complexity theory and thermodynamics authors like Bickhard (2000), Christensen and Hooker (2000), Collier (1999,2002), and Moreno and Ruiz-Mirazo (2000), have defended a more specific notion of autonomy as a recursively self-maintaining far-from-equilibrium and thermodynamically open system. The interactive side of autonomy is essential in the definition: autonomous systems must interact continuously to assure the necessary flow of matter and energy for their self-maintenance. The philosophical consequences derived from the nature of autonomous systems is highlighted by these authors, summarized in Collier's words: "No meaning without intention; no intention without function; no function without autonomy." (Collier, 1999). Autonomy is made the naturalized basis for functionality, intentionality, meaning and normativity1. But it is not always clear what the relation between basic (lower level thermodynamic) autonomy and adaptive behaviour is or how such thermodynamic processes interact with neural mechanisms. It is even argued that dynamical system theory (and thus computational simulation models) cannot capture the kind of organization that autonomy is or that robots should be self-constructive in order to be "truly" autonomous.

Key relevant Issues for Cognitive Science
In general and across the differences between the uses of the term autonomy and the consequences that (more or less explicitly) are derived from it, the notion of autonomy subsumes a set of key notions in cognitive science that have been pushing towards a paradigmatic shift. Among this key notions we find:

• Biological grounding: the idea that the understanding of cognition and adaptive behaviour must be approached bottom-up at two levels: in terms of the evolution of cognitive capacities in natural history and in terms of the biological mechanisms (neural networks, bodies, neuroendocrine systems, etc.) that produce cognitive behaviour.
• Distributedness, complexity, emergence: the idea that there is no central processor or homunculi that controlls behaviour but a distributed and functionally integrated network of recursive processes from which a coherent behaviour emerges as a global product of the system. The notion of autonomy assumes a high degree of complexity in the system introducing constraints on the possible analysis and functional localization and decomposition of structures in the system.
• Interactivism, embodiment, situatedness, dynamicism: Cognition is a process whose development and realization cannot be decoupled from the embodied interaction processes in which it is situated. An autonomous approach assumes a dialectics between independence and structural coupling; an interactive construction of meaning and behaviour in which embodiment and situatedness are taken to be essential features of cognition that are best captured by dynamical (rather than traditional computational) notions, thus introducing time and space dependant constraints as essential features on the generation of behaviour.
• Critique to symbolic approaches: The use of the notion of autonomy is often associated with a profound critique to what has been the mainstrain paradigm in cognitive science: the view that cognitive processes are logical transformations of computational states bearing a representational relation with observer independent "states of affairs" in the world; where the representational relation is taken to be the mark of the mental and the program-like transformation rules between representational states the causally efective mechanisms in the production of behaviour. From autonomous robotics to the philosophy of biology and cognition the approaches focused on autonomy have taken a different starting point, different theoretical primitives from which theories of cognition and adaptation have been built: complex dynamic networks, physically (and thermodynamically) embodied interactions, descentralized control systems, biologically grounded subsymbolic processes, etc. The theoretical development of autonomy as foundational concept in biology and cognitive science as well as the practice of autonomous robotics have chalenged profound theoretical assumptions of traditional approaches, directly denying the relevance of such assumptions (as it is the case of Prem with formalization and Smithers with information processing) or reformulating them in new grounds (as it is the case with Bickhard's notion of representation as indication of interaction outcome for the system or self-evaluable possibilities for action, or Varela's notion of in-formation). In any case cognitive functionality is never defined as mapping or correlation between internal and external states but always in a circular manner that has consequences for the self-maintenance of behavioural organization.

Current Research Areas Dealing with Cognitive Autonomy
A broad number of disciplines cover today the research field of autonomous system. Some of the most relevant are:

• Mathematical approaches to formalize the concept of autonomy: three main lines of research appear as the most relevant. Information theoretic approaches try to measure how much of the mutual information between system and environment is internally generated and controlled, hyperset and category theory applied to formalize circular organization and dynamical systems approaches studying spontaneous or self generated transitions and dynamic landscapes (e.g. chaotic itineracy).
• Origins of autonomous agents: searching for the minimal conditions for the appearance and sythesis of self-producing proto-cells ''in vivo'' and in ''silico'' and their agential (i.e. interactive regulatory) apacities shuch as ion-pumping, chemotaxis, etc.
• Biological grounding of cognition: how the nervous system is integrated withing bioregulatory processes and the role of emotion in cognitive regulation.
• Evolutionary robotics and minimal cognition: searching for minimal examples of neurodynamic embodied agents capable of cognitive development through situated action, internal and interactive homeostasis,
• Large scale neural dynamics: studying the holistic properties of neural dynamics, its homeodynamic and chaotic properties and its self-regulatory capacity and internal dynamic synchronization and coherency. Together with neurophenomenology (a bidirectional influence between first person conscious experience and large scale neuroscience).

References
Beer, R. D. (1997). The Dynamics of Adaptive Behavior: A research program. Robotics and Autonomous Systems, 20:257-289.

Brooks, R. A. (1991) Intelligence without representation. Artificial Intelligence Journal, 47:139-160.

Christensen, W. and Hooker, C. (2002). Self-directed agents. Contemporary Naturalist Theories of Evolution and Intentionality, Canadian Journal of Philosophy, 31 (special issue).

Collier, J. (2002) What is autonomy? In International Journal of Computing Anticipatory Systems: CASY 2001 - Fifth International Conference.

Di Paolo, E. (2003). Organismically inspired robotics. In Murase, K. and Asakura, T., editors, Dynamical Systems Approach to Embodiment and Sociality, pages 19-42. Advanced Knowledge International, Adelaide, Australia.

Maes, P., editor (1991). Designing Autonomous Agents. MIT Press.

Maturana, H. and Varela, F. (1980). Autopoiesis. The realization of the living. In Maturana, H. and Varela, F., editors, Autopoiesis and Cognition. The realization of the living, pages 73-138. D. Reidel Publishing Company, Dordrecht, Holland.

Prem, E. (1997). Epistemic autonomy in models of living systems. In Proceedings of the Fourth European Conference on Artificial Life. MIT Press, Bradford Books.

Prem, E. (2000). Changes of representational AI concepts induced by embobied autonomy. Communication and Cognition - Artificial Intelligence, 17(3-4):189-208. Special Issue on: The contribution of artificial life and the sciences of complexity to the understanding of autonomous systems. Guest Editors: Arantza Exteberria, Alvaro Moreno, Jon Umerez.

Ruiz-Mirazo, K. and Moreno, A. (2000). Searching for the Roots of Autonomy: the natural and artificial paradigms revisited. Artificial Intelligence, 17 (3-4) Special issue:209-228.

Smithers, T. (1995). Are autonomous systems information processing systems? In Steels, L. and Brooks, R., editors, The artificial life route to artificial intelligence: Building situated embodied agents. New Haven. Erlbaum.

Smithers, T. (1997). Autonomy in Robots and Other Agents. Brain and Cognition, (34):88-106.

Varela, F. (1979). Principles of Biologicall Autonomy. North-Holland, New York.

This briefing is based on an early draft of Barandiaran, X. (2004) Behavioral Adaptive Autonomy. A milestone in the Alife route to AI? Proceedings of the 9th International Conference on Artificial Life. MIT Press, Cambridge: MA, pp. 514--521.