Contributed by: Jan Wessnitzer , University of Edinburgh

Introduction
The field of Autonomous Robotics aims to build robotic systems capable of operating without external (human) control in complex environments by using situated approaches to cognition. Closely related to notions and studies of cognition, self-organisation and autonomy, Biorobotics is a new multidisciplinary subfield of robotics which aims to advance engineering and biology by mimicing and understanding biological systems in their ecological niches, drawing on neuroscience and studies of animal behaviour.

Nature and animals have long served as inspiration to engineering disciplines (e.g., [2]). Like animals, ultimately, a robot must be autonomous (function without any [human] supervision), be able to negotiate complex environments and to learn from its own experiences. As yet, animals are far more complex than any artificial system built. A truly biomimetic robot would imitate (proto-)cognitive processes, body structure, senses, and forms of locomotion from animals (ranging from invertebrates to vertebrates; such as insects or humans respectively). In this way, biorobotics inspires itself from evolutionary success (i.e., living organisms) as animals survive in unpredictable real world environments and exhibit far superior capabilities than robots in the typically controlled surroundings of a laboratory. With the goal of advancing engineering, robotics engineers use expertise from the fields of biology, computer science and engineering. Although biological research has made substantial advances regarding the brain functions of animals (at various levels - behavioural, physiological, etc.) and increases in computing power mean that neurophysiological brain functions can be mimicked in real time, current biological and technological knowledge is still far from creating artificial creatures that not only behave but also function in the same ways animals do.

Biorobotics not only aims to build better, more intelligent robots but it also aims to be a tool for biology (i.e., to test biological hypotheses, e.g., to develop algorithmic descriptions of brain functions) to further the understanding of the biological underpinnings by building robots that behave like animals. Thus, Biorobotics should be seen as having a dual purpose.

Modelling approach and methodology
Biorobotics approaches the understanding of cognition and adaptive behaviour by building complete systems; this means that an animal's or a robot's internal processes cannot be decoupled from the embodied interactions with the environment in which the animal or robot is situated. Useful properties of biological systems are stable yet adaptive interactions with the environment. Reverse engineering (tracking a result through its process to its source) has as a tenet that the cause exists. For example, just knowing there is an animal that can track moving objects while flying through space without visible light, proves that it is possible. Using hypotheses on how such behaviour can be achieved, it should be possible to formalise models and build robots achieving the same task.

Animals sense their environments with a wide variety of sensory receptors and organs. When using computer simulations, it is often necessary to include and simulate some properties of the real world as it is an important part of the complete system. Using robots has the advantage that the world in which it acts does not need to be simulated, thus preventing researchers from making unrealistic assumptions about the environment which could mislead researchers about the mechanisms required to deal with such an environment. Thus, a biorobotic approach emphasises the contribution of the physics of environments, sensors and actuators to the control of behaviour and allows testing models of (neural) systems within those constraints. Formalising models, besides producing testable predictions, may reveal that the hypotheses are unable to account for the experimental data.

As Biorobotics is an emerging field, no definite methodology yet exists. Research in Biorobotics should therefore clearly state what question motivated the research and how this question was answered in order to become a useful tool for biology. Some dimensions on which simulation models can differ and on which modelling decisions should be made were proposed in [10]:

1. Relevance: whether the model tests and generates hypotheses applicable to biology. A model is biologically irrelevant if it does not generate testable predictions but may still be useful for control, education, etc.
2. Medium: the physical basis by which the model is implemented. Physical implementations operate in environmental conditions with the constraints and the opportunities that such environments provide and using those not only saves efforts in modelling but also allows for validation.
3. Performance match: to what extent the model behaviour matches the target behaviour. Miniaturisation of robots means that experiments can be repeated with the same stimuli for the robot and the animal to gauge and quantify the behavioural match.
4. Structural accuracy: how well the model represents the actual mechanisms underlying the behaviour. Building multiple models of the same system is a useful strategy because results common to all these models compensate for the individual inaccuracies of each.
5. Level: the elemental units of the model in the hierarchy from atoms to societies. The level of the modelled mechanism reflects the level of information available from biology.
6. Abstraction: the complexity, relative to the target, or amount of detail included in the model. A more detailed model is less abstract. Assumptions involved in the abstraction should be clearly justified as details (abstracted away) may turn out to be critical.
7. Generality: the range of biological systems the model can represent. Can general principles emerge from modelling specific systems or do they only apply to single species, etc.

The biological plausibility of robot models may vary according to these dimensions and are discussed in depth in [10].

Examples of current research
To date, Biorobotics has been most successful in studying single sensorimotor systems, particularly in invertebrates. A number of successful robot implementations have been modelled on the sensorimotor control strategies of insects. This section provides (non-exhaustive and non-representative) examples of research in Biorobotics and discusses some examples briefly in terms of the modelling dimensions as outlined previously.

Subject area	Example	References
Simple sensorimotor control	Visual control
	-looming response	Rind, 2005 [8]
	-collision avoidance	Franceschini et al., 1992 [4]
	Taxis
	-phonotaxis	Webb, 1995 [9]
	-chemotaxis	Grasso et al., 1998 [5]
Locomotion
	Six-legged walking
	-stick insect	Duerr et al., 2003 [3]
	-cockroach	Quinn and Ritzmann, 2001 [7]
Navigation	Insect polarisation vision	Lambrinos et al., 2000 [6]
	Hippocampus	Burgess et al., 1997 [1]

For instance, simple sensorimotor strategies may lack in structural accuracy as these implementations often use wheeled rather than legged locomotion. However, despite this, a good behavioural match is achieved as the systems are embedded in real environments and subjected to the same sensory stimuli as the insect. Biorobotics research is concerned with complete systems, even if this sometimes limits the individual accuracy of particular subsystems. While undoubtedly greater accuracy is considered worth striving for, a degree of approximation is considered a price worth paying for the benefits of gaining a more integrated understanding of the system in a real environment. However, just because some robot has legs, it is not necessarily a biorobotic model of an animal. For example, military mule applications do not aim to be biologically relevant.

This synergy between biology and engineering when applied to more complex systems will however play a significant role in the advancement of (neuro-)ethology and of engineering disciplines. Models can be compared with respect to these proposed dimensions but should not be seen as way of ranking models, but rather as an elucidation of potential strategies [10].

References

1. Burgess, N., Donnett, J., Jeffery, K., and O'Keefe, J. (1997). Robotic and neuronal simulation of the hippocampus and rat navigation. Philosophical Transactions of the Royal Society B, 352:1535-1543.
2. De Solla Price, D. (1964). Automata in history: Automata and the origins of mechanism and mechanistic philosophy. Technology and Culture, 5(1):9-23.
3. Duerr, V., Krause, A., Schmitz, J., and Cruse, H. (2003). Neuroethological concepts and their transfer to walking machines. International Journal of Robotics Research, 22:151-167.
4. Franceschini, N., Pichon, J., and Blanes, C. (1992). From insect vision to robot vision. Philosophical Transactions: Biological Sciences, 337(1281):283-294.
5. Grasso, F., Basil, J., and Atema, J. (1998). Toward the convergence: robot and lobster perspectives of tracking odors to their source in the turbulent marine environment. In Proceedings of the IEEE International Symposium on Intelligent Control, pages 259-264.
6. Lambrinos, D., Moeller, R., Labhart, T., Pfeifer, R., and Wehner, R. (2000). A mobile robot employing insect strategies for navigation. Robotics and Autonomous Systems, 30:39-64.
7. Quinn, R. and Ritzmann, R. (2001). Biorobotics - methods and applications, chapter Construction of a hexapod robot with cockroach kinematics benefits both robotics and biology, pages 87-105. AAAI Press / The MIT Press.
8. Rind, F. (2005). Methods in Insect Sensory Neuroscience, chapter Bioinspired sensors: from insect eyes to robot vision. CRC Press.
9. Webb, B. (1995). Using robots to model animals: a cricket test. Robotics and Autonomous Systems, 16:117-134.
10. Webb, B. (2001). Can robots make good models of biological behaviour? Behavioral and Brain Sciences, 24:1033-1050.