Animat

From Scholarpedia
Jean-Arcady Meyer and Stewart W. Wilson (2009), Scholarpedia, 4(5):1533. doi:10.4249/scholarpedia.1533 revision #90965 [link to/cite this article]
Jump to: navigation, search

An animat is a simulated animal or a robot whose structure and functionalities are as much inspired from current biological knowledge as possible, in the hope that it will exhibit at least some of the survival capacities of real animals. The term "animat" was introduced in Wilson (1985).

Contents

The animat approach

Ever since the Dartmouth College workshop that founded the field of artificial intelligence in 1956 --- and the enthusiastic comments on the prospects for the discipline that this event triggered --- serious doubts (Dreyfus, 1972, 1992) have been raised about the chances that a computer program might compete in the near future with the amazing capacities exhibited by the human brain. In particular, several researchers (Brooks, 1991; Wilson,1991; Roitblat, 1995; Meyer,1995) consider that it is quite premature trying to understand and reproduce human intelligence --- whatever this expression really means --- and that one should first try to understand and reproduce the probable roots of this intelligence, i.e., the basic adaptive capacities of animals that have to deal with the full complexity of autonomous interaction with their environment. In other words, before attempting to reproduce unique capacities that characterize man, such as logical reasoning or natural language understanding, it might be wise to concentrate first on simpler abilities that human beings share with other animals, like navigating, seeking food and avoiding dangers. Similarly, Daniel Dennett (1978) posed the question "Why not the whole iguana?", arguing that the best route to approaching the complexities of human intelligence is to build simple, but complete creatures first.

The animat approach emphasizes characteristics neglected by standard AI approaches. In particular, it addresses problems of perception, categorization, and sensori-motor control and acknowledges that an organism doesn't always solve its survival problems by computational means and control algorithms. More generally, stressing the necessity of integrating both body - including its emotional and bioregulatory mechanisms - and control in the quest for understanding adaptive behaviors in natural or artificial systems, this approach is embodied and situated. It also contributes to foundational artificial intelligence. In this perspective, two key elements are, first, to take inspiration from what is known about the mechanisms underlying biological intelligence, and, second, to approach AI in a bottom-up manner, i.e. to go from simple, but ‘complete’, creatures to more complex ones.

The animat field received initial recognition on the occasion of the first SAB (Simulation of Adaptive Behavior) conference, which was held in Paris in September 1990. Since then, sub­sequent SAB conferences have been held every two years in Europe, US and Japan. Meanwhile, the quarterly journal Adaptive Behavior was launched in 1992, and The International Society for Adaptive Behavior (ISAB) was established in 1995.

Figure 1: The overall organization of an animat. Its control architecture decides which action to perform -- via its actuators -- depending on its current perceptions -- mediated by its sensors. Such permanent sensory-motor loops afford short-term adaptation with respect to a possibly changing, unpredictable and threatening environment. If the animat's viability zone is defined by two essential variables, V1 and V2, its adaptive behavior is evidenced by the corrective action it takes at point A to avoid crossing the viability zone at point B. Longer-term adaptive capacities are afforded by processes like learning, evolution, or development.

An animat's organization

An animat is usually equipped with sensors, actuators, and a behavioral control architecture that relates its perceptions to its actions and allows it to survive in its environment. In this context, survival depends upon some essential variables that must be monitored and maintained within a given viability zone (Ashby, 1952), an ability that can be enhanced by adaptive processes such as learning, development or evolution ( Figure 1).

Every aspect of Figure 1 is addressed by current research in animats (Meyer and Guillot, 2008).

Sensors

Numerous bio-inspired sensors have been implemented in animats. For example, the principles of optic-flow monitoring in the compound eye of the housefly are used for terrain-following, take-off or landing in a tethered flying engine (Ruffier et Franceschini, 2005) (Figure 2). Likewise, other sensors in animats are inspired by audition, touch, smell, or taste devices invented by nature. The general trend underlying these efforts is to delegate low-level processes to dedicated upstream devices, thus simplifying the task of the control architecture, which may then focus on the treatment of high-level information.

Figure 2: An opto-electronic device inspired by the housefly’s compound eye affording a flying animat terrain-following, take-off and landing capacities.(c) CNRS Photothèque, Nicolas Franceschini, UMR6152 - Mouvement et Perception Marseille

Actuators

Bio-inspired solutions affording crawling, walking, jumping, swimming, flying, grasping, and drilling capacities have been implemented in animats. For instance, to make wall-climbing possible for a gecko-like robot, its feet are covered with thousands of synthetic setae made of an elastomer that ensure a large area of contact between the feet and the wall, thus maximizing the expression of intermolecular van der Waals forces (Santos et al., 2008) (Figure 3). Again, the general trend is to use materials whose mechanical properties simplify the associated control problem.

Figure 3: Stickybot, an artificial gecko. (c) Mark R. Cutkovsky, Stanford University, Center for Design Research

Control architectures

To allow them to "survive" or fulfill their mission in unpredictable environments, the control architectures of animats often implement, beyond mere reflexes connecting their sensors to their actuators, a variety of other adaptive properties or processes such as active perception, efferent copies, motivations, emotions and action selection (Guillot and Meyer, 2003).

Often, these properties or processes serve to cope with present circumstances only. This is why more-cognitive architectures, able to deal with past and future events as well, are endowed with memory and planning capacities. Moreover, other adaptive mechanisms, like learning, evolution and development, are also currently incorporated into them.

Likewise, whereas the majority of these research efforts are targeted at increasing an animat’s behavioral autonomy, a new trend emerges that aims at tackling the challenge of reproducing the energetic autonomy of animals The robot EcoBot II, for instance, is equipped with microbial fuel cells, in the anodes of which bacteria found in sludge act as catalysts to generate energy from dead flies supplied by a human operator or autonomously caught by the robot (Ieropoulos et al., In press) (Figure 4). Studying the way animals manage to discover and exploit resources supplying their energy needs, one may hope to understand how the behavioral and energetic autonomy constraints interact in animals and impact their cognitive development and capacities.

Figure 4: EcoBot-II, an energetically autonomous robot. (c) Chris Melhuish, University of Bristol and University of the West of England

A specific animat: Psikharpax

The Psikharpax project (Meyer et al., 2005) is currently sponsored by the European integrated project ICEA (Integrating Cognition, Emotion and Autonomy - http://www2.his.se/icea/) and aims at producing an artificial rat equipped with control architectures and mechanisms that reproduce as nearly as possible those that have been widely studied in the natural rat. Psikharpax (also called ICEAbot) is a 50 cm-long robot (Figure 5) equipped with three sets of allothetic sensors: a two-eyed visual system, an auditory system calling upon two electronic cochleas, and a haptic system made of two whisker arrays on each side of its head. It is also provided with three sets of idiothetic sensors: a vestibular system reacting to linear and angular accelerations of its head, an odometry system monitoring the length of its displacements, and capacities to assess its current energy level. It also exhibits internal needs--such as hunger, rest, or curiosity. Later on, it will be endowed with background emotions--such as calm, tension, or well-being-- and primary emotions--like fear, disgust, or surprise.

Figure 5: Psikharpax, an artificial rat. (c) Christophe Grand, Steve Nguyen and Patrick Pirim, Institut des Systèmes Intelligents et de Robotique, UMR 7222, Université Pierre et Marie Curie, Paris

Its control architecture is inspired from the anatomy and physiology of dedicated structures in the rat’s brain, like the hippocampus, the basal ganglia and the cortex which afford navigation, action selection and planning capacities. Among the various learning mechanisms that are implemented in this architecture, a variety of reinforcement learning calling upon dopamine signals in the basal ganglia (Khamassi et al., 2005) makes it possible to reach rewarding goals and to avoid nociceptive objects.

Video of Psikharpax (2009)


The adaptive capacities of Psikharpax were demonstrated through simulations (Girard et al., 2005) in which the animat was equipped with a simplified sensory equipment, such as a visual system sampling the mean grey-level of the environment in 36 directions. In particular, it was shown that Psikharpax is able to "survive" by permanently maintaining its internal energy level above a viability threshold. This entails finding a given resource in the environment, "eating" it and, then, returning to the "nest" to "digest" it. This also entails exploring the environment, and building a "cognitive map" that serves to position itself and to locate both the nest and the resource. Finally, this entails being able to select at every moment an appropriate navigation strategy, such as directly aiming toward a visible goal, or using the map to navigate toward a remembered one while avoiding passing through dangerous places.

However, the benefits of using a truly rich sensori-motor equipment can be investigated only through real-world experiments, because no simulation is accurate enough to correctly reproduce the inner workings of, for example, a whisker system. In addition, only through real-world experiments is it possible to assess how a given sensory modality may compensate for the absence or the defect of another and to demonstrate that some of the robot's adaptive capacities are maintained even in the dark, i.e., when, in effect, touch and audition jointly supply vision.

Thus it appears that, while simulations and robotic implementations have often been opposed in the past, the animat approach capitalizes on both, using each to best advantage.

References

  • Ashby, W.R. (1952). Design for a Brain, Chapman & Hall.
  • Brooks, R.A. (1991). Intelligence without Representation. Artificial Intelligence 47, 139-159.
  • Dennett, D.C. (1978). Why not the Whole Iguana? Behavioral and Brain Sciences 1:103-104.
  • Dreyfus, H.L. (1972). What computers can’t do: The limits of Artificial Intelligence, Harper and Row.
  • Dreyfus, H.L. (1992). What computers still can’t do: A critique of Artificial Reason, The MIT Press.
  • Girard, B., Filliat, D., Meyer, J.-A., Berthoz, A., and Guillot, A. (2005). Integration of navigation and action selection functionalities in a computational model of cortico-basal ganglia-thalamo-cortical loops. Adaptive Behavior, 13(2):115-130.
  • Guillot, A. and Meyer, J.-A. (2003a). La contribution de l'approche animat aux sciences cognitives. In Cognito, 1(1):1-26.
  • Ieropoulos, I., Melhuish, C., Greenman, J. and Horsfield, I. (In press). Artificial symbiosis in EcoBots-I and II. In Adamatzky, A. and Komosinski, M. (Eds). Artificial Life Models in Hardware.
  • Khamassi, M., Lachèze, L., Girard, B., Berthoz, A., and Guillot, A. (2005). Actor-critic models of reinforcement learning in the basal ganglia: From natural to artificial rats. Adaptive Behavior, Special Issue Towards Artificial Rodents, 13(2):131-148.
  • Meyer, J.A. (1995). The Animat Approach to Cognitive Science. In: Roitblat, H.L., Meyer, J.A. (Eds.), Comparative Approaches to Cognitive Science. The MIT Press, pp. 27-44.
  • Meyer, J.-A., Guillot, A., Girard, B., Khamassi, M., Pirim, P., and Berthoz, A. (2005). The Psikharpax project: Towards building an artificial rat. Robotics and Autonomous Systems, 50(4):211-223.
  • Meyer, J.-A. and Guillot, A. (2008). Biologically-inspired robots. In Siciliano, B. and Khatib, O., editors, Handbook of Robotics. Springer-Verlag.
  • Roitblat, H.L. (1995). Comparative Approaches to Cognitive Science. In: Roitblat, H.L., Meyer, J.A. (Eds.), Comparative Approaches to Cognitive Science. The MIT Press, pp. 13-26.
  • Ruffier, F. and Franceschini, N. (2005). Optic flow regulation: the key to aircraft automatic guidance. Robotics and Autonomous Systems. 50 (4), 177-194.
  • Santos, D., Heyneman, B. Kim, S. Esparza, N. and Cutkosky, M.R. (2008). Gecko-Inspired Climbing Behaviors on Vertical and Overhanging Surfaces. Proceedings IEEE ICRA 2008, Pasadena, May 19-23.
  • Wilson, S.W. (1985). Knowledge growth in an artificial animal. In: Grefenstette, J.J. (Ed.), Proceedings of the First International Conference on Genetic Algorithms and Their Applications (pp. 16-23). Hillsdale, New Jersey: Lawrence Erlbaum Associates.
  • Wilson, S.W. (1991). The Animat Path to AI. In: Meyer, J.A., Wilson, S.W. (Eds.), From animals to animats: Proceedings of the First International Conference on Simulation of Adaptive Behavior. The MIT Press, pp. 15-21.

Internal references

  • Joseph E. LeDoux (2008) Amygdala. Scholarpedia, 3(4):2698.
  • Valentino Braitenberg (2007) Brain. Scholarpedia, 2(11):2918.
  • Olaf Sporns (2007) Complexity. Scholarpedia, 2(10):1623.
  • Edvard Moser and May-Britt Moser (2007) Grid cells. Scholarpedia, 2(7):3394.
  • Howard Eichenbaum (2008) Memory. Scholarpedia, 3(3):1747.
  • Giacomo Rizzolatti and Maddalena Fabbri Destro (2008) Mirror neurons. Scholarpedia, 3(1):2055.
  • Wolfram Schultz (2007) Reward. Scholarpedia, 2(3):1652.
  • Kathleen Cullen and Soroush Sadeghi (2008) Vestibular system. Scholarpedia, 3(1):3013.


Recommended reading

  • Arkin, R. (1998). Behavior-Based Robotics. The MIT Press.
  • Brooks, R. (2003). Flesh and Machines: How Robots Will Change Us. Vintage.
  • Floreano, D. and Mattiussi, C. (2008). Bio-Inspired Artificial Intelligence: Theories, Methods, and Technologies. The MIT Press.
  • Nolfi, S. and Floreano, D. (2000). Evolutionary Robotics: The Biology, Intelligence, and Technology of Self-Organizing Machines. The MIT Press.
  • Pfeifer, R. and Scheier, C. (2001). Understanding Intelligence. The MIT Press.
  • Pfeifer, R. and Bongard, J. (2006). How the body shapes the way we think – a new view of intelligence. The MIT Press.
  • Roitblat, H. and Meyer, J.-A., editors, (1995). Comparative Approaches to Cognitive Science. The MIT Press.
  • Steels, L. and Brooks, R., editors. (1995). The Artificial Life Route to Artificial Intelligence. Building Embodied, Situated Agents. Lawrence Erlbaum.
  • Proceedings of the SAB Conferences (1990-2008): From Animals to Animats.

External links

See also

Action selection, Amygdala, Cognition and emotion, Computational models of classical conditioning, Developmental_robotics, Evolutionary algorithms, Evolutionary_robotics, Grid cells, Head direction cells, Mirror neurons, Models of basal ganglia, Models of emotion, Models of mirror system, Neurorobotics, Reinforcement learning

Personal tools
Namespaces

Variants
Actions
Navigation
Focal areas
Activity
Tools