Morphological computation (robotics)

Morphological computation is computation obtained through interactions of physical form.

Birth of the term morphological computation

The term morphological computation was first coined by Chandana Paul in the context of legged robotics. In work on biped robots, Paul had observed that smooth variation of single parameters of an open loop controller could give rise to discrete changes in limit cycle behavior. Although there was no discrete change specified in the control, the interaction of the controller with the morphology produced a discrete change in behavior, almost as if a bit were suddenly being flipped in the controller. This puzzling phenomenon led Paul to question whether it may be possible for the morphology, or the physical body of the robot, to perform computation.

In August 2002, Paul was reading a book which gave a simple explanation about how perceptron networks could not give rise to functions which were non-linearly separable (note that this only applies to single-layer perceptrons, that is, linear combinations of inputs. a multilayer perceptron can theoretically approximate any function). This simple fact had been well-established in the past, but it now raised a new doubt. Paul felt that if the morphology of a robot could indeed perform computation, then it should be possible to design a robot which used perceptron networks to perform a non-linearly separable function.

She set herself the challenge to design such a robot. After many trials and design iterations, Paul finally discovered the design of the XOR robot. This robot, elegant in its simplicity with only two moving parts, could use perceptron networks to display the non-linearly separable function XOR, in its behavior. The robot represented the first proof that physical interactions in robot morphology could give rise to computation. Paul coined the term "morphological computation" to describe this phenomenon.^[1]

In 2009, Jordi Vallverdú, with SETE research group (UAB), made a XOR robot with Lego Mindstorms using the AND + OR approach, but at the same time created a more simple way to obtain a XOR in this kind of robots. Their results were presented at EUCogII conference (Hamburg, October 10-11th, 2009).

Relevance

The understanding that simple physical interactions can give rise to computation has far-reaching implications in many domains, which are only beginning to be understood.

Robotics

The understanding that the morphology performs computation can also be used to analyze the relationship between morphology and control and to theoretically guide the design of robots with reduced control requirements.^[2]

This provides the basis for a new field of research, in which theoretical methods can be developed to analyze and quantify the relationship between morphology and control ^[3][4]

Based on such analysis, robots can be designed which use the morphology to perform a computational role in the system, and the ease the computational requirements on the controller. A nice example of this was provided by Rolf Pfeifer and Fumiya Iida, introducing the idea to the robotics community, through the design of a robotic hand.^[5].

Note: Some people in robotics use the phrase "morphological computation" in a broad sense as describing the trade-off between morphology and control in robots. That is, every instance in which the robot morphology is shown to aid the control, they say the body performs "morphological computation". This is technically inaccurate, as there are other mechanisms besides computation which can also give rise to the morphology and control trade-off. Without an analysis which proves the existence of computation in the specific robot structure, it is not accurate to claim that it performs morphological computation.^{[citation needed]}

Artificial intelligence

The understanding that simple physical interactions give rise to computation opens the door to the possibility that the physical interaction of a robot body with the environment can also be used to enhance its computational abilities. In essence the world can become the "abacus" for the computations of the intelligent agent.^[1]

Cognitive science

The understanding that the physical body and its interaction with the environment can enhance computational abilities, also opens the door to a new field of investigation in cognitive behavior, into the way humans and animals exploit the environment in their behavior to enhance computation.

One example, of this is seen in counting in humans. When there are a very large pile of coins for example that need to be counted on a table, a person does not rely on pure computational processing of visual input to accomplish the task. Instead they use their hand to physically move the coins one by one from one pile to another all the while incrementing a mental counter. Thus, they utilize the physical interaction with the environment to enhance their computational ability and accomplish an otherwise computationally intractable task.^[1]

The relevance of morphological computation in adaptive behavior and cognition has been more extensively demonstrated with various examples of robots interacting in the real world ^[6]

Molecular processes

The only significant physical characteristics of Paul's XOR robot which are necessary to give rise to computation are the interaction between moving parts, and the occurrence of discrete events. Both these features are also found in molecular interactions, and thus it has been proposed that morphological computation can also occur in simple molecular interactions.^[2]

Biomechanics

It has been found that the musculo-skeletal systems of biological organisms also embody the necessary characteristics to enable computation.^[2] This understanding could lead to new roads of investigation into how evolution has optimized musculo-skeletal systems to enable computations which help motor control.

Thus, the study of morphological computation is relevant to many areas of science and engineering and will serve to conceptually unify the investigation of many natural and artificial domains.

Events

The first international event on the topic of morphological computation in robotics was the International Workshop on Morphology, Control and Passive Dynamics, held at the Intelligent Robots and Systems Conference (IROS) in Edmonton, Canada, June 2005. This was organized by Chandana Paul, Andy Ruina, Max Lungarella, Manoj Srinivasan, and Fumiya Iida, and addressed the role of morphological computation in robot control, particularly in underactuated motion ^[3].

Following this, the First International Conference on Morphological Computation was organized by Rolf Pfeifer, Norman Packard, Mark Bedeau, and Fumiya Iida in Venice in March 2007. This was a more general event covering morphological computation in a wider range of areas such as robotics, amorphous computing, organic computing, membrane computing, DNA-computing, self-assembled computing and molecular computing.^[7].

The second International Conference on Morphological Computation (ICMC2011) will take place in Venice for 12-14th September 2011. The mission of this conference is to bring the leading researchers from wider range of research areas including several growing areas such as reservoir computing and soft robotics in order to address the formulation and development of a new paradigm for fully embodied computation.

References

1. ^ ^{a b c} C. Paul (2004) Morphology and Computation, Proceedings of the International Conference on the Simulation of Adaptive Behaviour Los Angeles, CA, USA, pp 33–38
2. ^ ^{a b c} C. Paul (2006) Morphological Computation, Robotics and Autonomous Systems, Special Issue on Morphology, Control and Passive Dynamics, August 2006, 54(8): 619–630
3. ^ ^{a b} C. Paul, A. Ruina, M. Lungarella, M. Srinivasan, F. Iida (2005) Proceedings of the Workshop on Morphology, Control and Passive Dynamics, August 2005, CD-ROM publication.
4. ^ C. Paul, M. Lungarella, F. Iida (2006) Morphology, Control and Passive Dynamics, Special Issue of journal Robotics and Autonomous Systems, Vol. 54(8), August 2006
5. ^ R. Pfeifer and F. Iida (2005). Morphological computation: Connecting body, brain and environment. Japanese Scientific Monthly, Vol. 58, No. 2, 48–54
6. ^ R. Pfeifer and F. Iida. Morphological Computation for Adaptive Behavior and Cognition. International Congress Series. (2006) 1291: 22–29
7. ^ R. Pfeifer, N. Packard, M. Bedeau, F. Iida. (2007) Proceedings of the International Conference on Morphological Computation, March 2007, CD-ROM.