CICADA aims to increase significantly the advancements of biomimetic life-like perception systems by providing novel data and concepts on a 'sensing-perception-action' chain. The specific objectives are the integration of knowledge from the biological paradigm (the flow of information along the chain for crickets avoiding danger) with the technological conceptualisation.

• Quantify the chain, from the stimulus, air velocity, to the escape response (IRBI)
• Abstract the information to provide the necessary technological developments and integration of biological sensors (UREADEN)
• Develop the "hardware", "software", and "wetware" for implementation in MEMS-based arrays of sensors (UT-MESA)
• Build up a hybrid demonstrator by combining the sensor output and the stimulation input to the neuro-electronic device and using the neural network to process information (FZJ)

Escape response in crickets (IRBI)

Escape orientation in arthropods is a highly stereotyped, quick and precise behavior based on a proper recognition of stimuli announcing imminent danger from a predator attack. The escape behavior is triggered by physical stimuli which need to be detected (air-flow, pressure, sound, vibrations, mechanical forces,…), locally amplified and filtered by specialized sense organs, transmitted, processed and integrated into a perception pattern which will control response. For preys, the sensory apparatus as well as the neural mechanisms which have been studied in greatest detail occur in crickets, for which we have a good understanding at the cellular and neuronal level. The stimuli resulting escape reactions in these cases are known to be air currents produced by an approaching predator, being a flying or running wasp or the tongue of a toad. The results obtained so far seem to apply also fairly well to the escape mechanisms of cockroaches and have strong similarities to the startling mechanisms of caterpillars and crayfishes. Thus, while we will use preferentially one species of cricket, we expect our results to be of wide relevance for many other insect groups and even for crustaceans living in water.

By contrast to the large number of physiologically and neuro-ethological oriented studies, we know almost nothing about the selection forces having led to the appearance and conservation of this sensory and perception system. In particular, we do not known which are the main predators, in the field, of these crickets. Hence, we do not know much about the form of the stimuli and their time/space representation both in physical terms and as perceived by the prey. These biologically oriented questions urgently require answer even from the technology point of view, as selection pressures determine the degree of precision of the sensors, danger perception and escape reactions of the preys.

Biological sensors (UREADEN)

Two main types of sense organs are found in arthropods which react to the external physical quantity, thus implementing the first step of information transmission: hairs on the exoskeleton (hair sensilla) and modified structures within the exoskeleton itself, campaniform sensilla. Hair sensilla (sometimes in combination with campaniform receptors) can detect air-flow and signal the presence of predators. In the case of hair sensilla, depending on the shape of the hairs and the specific interaction of the hairs with underlying dendrites, the hairs can serve as sensing elements for physical stimuli such as air-flow, sound, temperature, humidity, mechanical forces and more. Yet, it appears that a combination of hairs, delicately distributed over certain areas, allows insects to distinguish between these differing effects. Undoubtedly the intricate combination of the stimuli of multiple hairs help the insects to establish this discriminating perception. This sensory apparatus is made of an array of filiform hairs of different length whose mechanical properties have been studied in detail, both theoretically and experimentally (structure, response function of single hairs). These hairs often sit on specialized structures (cerci in crickets). They are the most sensitive sensors known, even better than the best photoreceptors.

The hairs are usually innervated by a single cell, which is a simple system compared to many other sensillae. In order to transmit information effectively, the axons of some of these hairs are large, bundled together from one cercus, reaching as quickly as possible the terminal ganglion located peripherally in the abdomen. The next level of integration is the 6th abdominal ganglion which results from the fusion of the 5 most posterior ganglia. Ganglia are where the danger perception and a part of the reaction takes place. In particular, neural control of the first part of the turning away behavior in cockroaches is not centrally controlled but at the level of ganglia.

The working of many hairs, being in arrays or not, is not understood. It has been postulated that the coherence of hair movement and the firing of their neurons would enable insects to perceive the slightest air movements. In particular, it has been shown/postulated that a single hair is working at the thermal noise level, but that an array is able to filter this noise away (stochastic resonance has been postulated). Thus the arrays are able to pick up signals buried in noise while false alarm are kept to a low occurrence. The number of hairs varies a greatdeal species, up to several hundreds and thousands in crickets. The cercal inputs are afferents from altogether about 1000 filiform hairs, 1200 campaniform in one cricket species, and more in other species.

Since the overall sensing-perception-action system starts with the sensors itself one cannot come past the need to devote part of the work to exploration and fabrication of sensor structures. However, where “classical” sensor development is aimed at high specificity and low cross sensitivity, in this project the emphasis is on integrability and the interplay between sensor- and array-properties. This also means that cross-sensitivity in this project is welcomed as a “natural” condition found in biological systems. In some insects, for example, infrared sensors appear to have evolved from sense organs designed to detect mechanical deformation. Since there is an equivalence between thermally and mechanically induced deformations, optimisation strategies in the design of the sense organ or in the processing of information are necessary to avoid unwanted cross-coupling effects.

MEMS (UT-MESA)

In order to partly mimic and partly investigate the interaction between sensing hairs and the connected neuronal systems on the perception systems of crickets we will fabricate (large) arrays of mechanical sensors by means of MEMS technology. In doing so we will investigate the influence of placement, collectivity, dissimilarity of hairs and cross-sensitivity on sensing selectivity. At this stage, the important trade-off between the sensor sensitivity and robustness against other quantities will have to be analyzed. Moreover and additionally, we will investigate and where possible mimic the potential perception stages existing in the various neuronal levels. We are aware that this requires the challenging identification of these stages as well as the level of abstractness existent at each stage. By making an artificial sensing and neural system we hope to be able to explore the structure, the variety, the complexity and accuracy of the sensing system, and the potential levels of hierarchy and abstractness in perception. Additionally we hope to be able to synthesize a system that enables to mimic, or even verify, some of the findings of the functioning of the sensing-perception system of insects. Array topology and technology will also be investigated to explore implementation strategies for global sensing, localized action and manufacture of sensor-arrays on flexible media. Biological systems sometimes show large reductions in number of connections “going upward” to the brain. Such large reductions are presumably possible due to either redundancy or smart “sensor-fusion”. Additionally escape reactions are so fast that they can be named “localized actions”. It will be interesting to see if specific topologies are prerequisite to enable these kind of actions or if these actions are solely depending on the higher levels of the sensing-perception system.

In order to carry out the research proposed above we will make use of tiny flow-sensors and flow-sensor arrays. These structures will be fabricated by MEMS technology which allows us to make sensors and arrays comparable in size, density and variety to those found in the studied cricket species.

Information from large arrays of sensors need to be integrated and processed in order to convey simpler information used in the decision process. This is where going back to biology can help us tremendously. The overall goal is to determine how ensembles of neurons in a mapped sensory system extract information about dynamic stimuli and encode that information in their ensemble spike trains. The specific aims of the studies are to determine what parameters of sensory stimuli are encoded in the spike trains of the neurons; what is the nature of the neural encoding scheme with which that information is represented in the ensemble spike train patterns; and to examine the structural and biophysical mechanisms through which the observed coding scheme is implemented within this neural network. This knowledge is of particular interest for devising a viable sensory system architecture for the artificial MEMS based systems once they are available and the interfacing to neural cells is successful.

Hybrid demonstrator (FZJ)

The sensory system mediates the detection and analysis of low velocity air currents in the animal's immediate environment. Air currents deflect mechanosensory hairs. All relevant information about the direction and dynamics of the air current environment carried to higher centers is contained in the many spikes originating from a bundle of sensory cells. Most of previous studies on this neuronal integration have been carried out on single cells using intracellular microelectrodes. Although one can learn a great deal from these intracellular recordings about neural encoding at the single-cell level, one is not able to address more complex questions related to ensemble encoding. While neuroscientists have historically been limited to recording the activity of single cells for technical reasons, rapid developments on the technical side allows to record from several cells at the same time with high time resolution. Using multi-unit extracellular recording techniques one would aim to discriminate activity in individual neurons in a group of cells. The goals of the studies to be carried out will be to expand to a comprehensive analysis of dynamic encoding in larger ensembles of cells, through the use of multi-unit recording and analysis techniques.

When MEMS devices are to be connected to living biological cells, without doubt the materials required for doing so are different from the (technical) materials used normally in MEMS technology. Specifically inert and bio-compatible materials will be investigated, capable of providing the multifunctionality needed to perform mechanical and/or electrical functions.The greatest challenge so far has been therefore in building the interface between the living biological cell and technological hardware.

The proposed demonstrator, which will incorporate the collective knowledge generated by the project, is considered an effective mechanisms for stimulating and strengthening the multi-disciplinary elements of the proposal. Ideally, it should be made of a miniature sensing unit carrying potentially many MEMS sensors. The processing of the information from these arrays would be carried out in neuron cell cultures which would send information leading to a decision. The decision will be: no danger, do not move and danger, move in that direction. The demonstrator will however only virtually move since we do not anticipate the design of robotic structures.