Journal of Comparative Physiology A

, Volume 199, Issue 11, pp 1037–1052

Insect–machine hybrid system for understanding and evaluating sensory-motor control by sex pheromone in Bombyx mori


    • Research Center for Advanced Science and TechnologyThe University of Tokyo
  • Ryo Minegishi
    • Research Center for Advanced Science and TechnologyThe University of Tokyo
    • Department of Mechanical and Control EngineeringTokyo Institute of Technology
  • Shigehiro Namiki
    • Research Center for Advanced Science and TechnologyThe University of Tokyo
    • HHMI Janelia Farm Research Campus
  • Noriyasu Ando
    • Research Center for Advanced Science and TechnologyThe University of Tokyo

DOI: 10.1007/s00359-013-0832-8

Cite this article as:
Kanzaki, R., Minegishi, R., Namiki, S. et al. J Comp Physiol A (2013) 199: 1037. doi:10.1007/s00359-013-0832-8


To elucidate the dynamic information processing in a brain underlying adaptive behavior, it is necessary to understand the behavior and corresponding neural activities. This requires animals which have clear relationships between behavior and corresponding neural activities. Insects are precisely such animals and one of the adaptive behaviors of insects is high-accuracy odor source orientation. The most direct way to know the relationships between neural activity and behavior is by recording neural activities in a brain from freely behaving insects. There is also a method to give stimuli mimicking the natural environment to tethered insects allowing insects to walk or fly at the same position. In addition to these methods an ‘insect–machine hybrid system’ is proposed, which is another experimental system meeting the conditions necessary for approaching the dynamic processing in the brain of insects for generating adaptive behavior. This insect–machine hybrid system is an experimental system which has a mobile robot as its body. The robot is controlled by the insect through its behavior or the neural activities recorded from the brain. As we can arbitrarily control the motor output of the robot, we can intervene at the relationship between the insect and the environmental conditions.


BrainAdaptive behaviorMothsPheromonesOrientation



Antennal lobe


Brief excitation


Brain–machine hybrid system


Cervical nerve


Flip-flop activity


Delta area of inferior lateral protocerebrum


Lateral accessory lobe


Lateral protocerebrum


Macroglomerular complex


Neck motor neuron


Ordinary glomerulus


Olfactory receptor neuron




Projection neuron


Superior medial protocerebrum


Ventral protocerebrum


Adaptability to changing circumstances is a key feature of animals. Understanding adaptability of an animal behavior is one of the most important themes in the field of neuroethology. Animals living in the real environment perform appropriate behaviors according to the changing environmental conditions. In particular, insects are the most diverse and abundant animal group, representing more than 70 % of all known animal species. They reside almost everywhere on the earth and display a range of sophisticated behaviors based on the perception of a multitude of stimuli.

Here we define adaptability as the ability to execute a behavioral task under changing environmental conditions. This topic is of interest for robotics because robots also need to execute tasks in changing environmental conditions. The algorithm contained in the insect central nervous systems will lead us to new ideas for constructing algorithms for behaving properly under changing environment to complete a task.

Insects are valuable model systems in neuroethology due to the balance between the moderate complexity of their nervous systems and a rich behavioral repertoire. Insect brains contain on the order of 105 to 106 neurons, whose processing achieves the high performances of behavior under real circumstances (Mizunami et al. 1999, 2004). Insects are well suited for multidisciplinary studies in brain research involving a combined approach at various levels, from molecules to single neurons to neural networks, behavior, and modeling. These preparations are amenable to a wide variety of methodological approaches, in particular genetic engineering, neuroanatomy, electrophysiology, and functional imaging (Kanzaki et al. 2008).

Researchers have investigated the relationship between sensory inputs and behavioral outputs to find out the neural mechanisms underlying adaptive behavior using these approaches. However, since the adaptive behavior is generated through the interaction between the changing environment and insects, it is inadequate to study only static response properties of the nervous system by only giving constant stimuli. Therefore, some researchers have tried to record neural activities from freely moving insects (Mizunami et al. 1998; Okada et al. 1999; Diorio and Mavoori 2003). Other researches have established closed-loop experiments using virtual environments to mimic real environment stimuli (Gray et al. 2002; Reiser and Dickinson 2008; Ejaz et al. 2011).

In addition to these methods, there are some studies that refer to interface between insect and robots (Halloy et al. 2007; Melano 2011). In these studies, the insect-size robots were used to interact with groups of insects to control group behavior (Halloy et al. 2007), and the robot platform was used to record physiological signals of an insect (Melano 2011). Like these platforms, another experimental method is proposed, which meets the conditions necessary for approaching the dynamic processing in the brain of insects for generating adaptive behavior. We call this experimental method an ‘insect–machine hybrid system’ (Emoto et al. 2007; Minegishi et al. 2012; Ando et al. 2013). The insect–machine hybrid system is an experimental system which has a mobile robot as its body. The robot is controlled by an insect’s behavior (i.e., an insect-controlled robot) (Emoto et al. 2007; Ando et al. 2013) or the neural activities recorded from the brain (i.e., brain–machine hybrid system) (Minegishi et al. 2012). As we can arbitrarily control the motor output of the robot, we can alter the relationship between the insect and the environmental conditions (Fig. 1).
Fig. 1

Concept of the insect–machine hybrid system. Insect–machine hybrid system compared to that of a real organism from the viewpoint of how they process stimuli from the external environment

In this paper, focusing on sensory-motor control by sex pheromone in the silkmoth (Lepidoptera: Bombyx mori) we review the behavioral strategy for odor source orientation and its neural mechanisms as well as a novel approach using an insect–machine hybrid system for evaluating and understanding adaptive behavior in insects.

Odor source orientation behavior in the silkmoth as a model of adaptive behavior

One of the adaptive behaviors of insects is high-accuracy odor source orientation. Male moths exhibit mating behavior in response to sex pheromones emitted by conspecific females (Butenandt et al. 1959; Kennedy 1983; Vickers 2000; Cardé and Willis 2008). Inside a plume, pheromones do not form a continuous concentration gradient, but are discontinuously distributed (Murlis and Jones 1981; Murlis et al. 1992). The discontinuous portion of the pheromone is called the filament, and moths detect the filament as a pheromone input (Vickers and Baker 1994; Mafra-Neto and Cardé 1994). It is difficult to describe the diffusion dynamics of odor plumes in the real environment, because there are many factors which can make the behavior of plumes fluctuate in the real environment. In the field of engineering, a variety of chemical plume tracing robots implementing artificial algorithms have been suggested, but it is difficult to evaluate them (Ishida et al. 1996; Russell 2001; Kowadlo and Russell 2008). Since we do not know the real fluid dynamics, the functioning of the robots depends on how experimenters set the environmental conditions.

In flying moth species several candidate strategies for odor source localization have been proposed (Farkas and Shorey 1972; Kennedy 1940; Baker 1990; Mafra-Neto and Cardé 1994; Vickers and Baker 1994; Vergassola et al. 2007). In the optomotor anemotaxis hypothesis, male moths exhibit flight behavior in response to sex pheromone and the flight path is modified by visual and mechanical (i.e., air flow) feedback (Kennedy 1940; Kennedy and Marsh 1974; Baker et al. 1984). Further, it has been proposed that the moths intrinsically generate sustained flight during plume tracking (Baker and Kuenen 1982; Willis and Baker 1987; Willis and Arbas 1991; Vickers and Baker 1992). In addition to sustained flight, upwind surge flight in the initial phase of mating behavior has been observed in several moth species and is thought to have an important role in odor source localization (Baker 1990; Vickers and Baker 1992, 1994, 1996; Mafra-Neto and Cardé 1994, 1996). Thus in flying moths, a combination of optomotor anemotaxis and intrinsically programmed behavior, upwind surge flight at the initial phase and sustained zigzagging flight at the later phase are proposed (Baker 1990).

In contrast, male silkmoths exhibit a characteristic zigzagging pattern as they walk upwind toward the pheromones released by females of the same species (Fig. 2) (Butenandt et al. 1959; Kaissling 1971; Kramer 1975, Kramer 1986; Obara 1979; Kanzaki et al. 1992). Two components (bombykol and bombykal) of the pheromone have been identified and synthesized (Butenandt et al. 1959; Kaissling 1971). The major component, bombykol is sufficient to drive the complete zigzagging behavior for odor source localization (Kanzaki et al. 1992). Upwind walking toward a pheromone source is largely controlled by an internally generated steering program, triggered by the detection of an intermittent distribution of pheromones by the antennae. Once initiated by a single puff of pheromone, the moth exhibits a programmed sequence of walking consisting of brief bouts of straight-line walking (surge) and zigzag turns followed by looping (i.e., turns of more than 360°) (Fig. 2) (Kanzaki et al. 1992). Upon stimulation, male moths exhibit straight-line walking in the direction of the antenna to which the stimulation is applied (Takasaki et al. 2012). After the pheromone stimulation, males exhibit zigzagging walking with a significant increase in time between each turn followed by looping. This programmed sequence of movements is reset and restarted from the beginning in response to pulsed pheromonal stimulation.
Fig. 2

Male silkmoth and programmed behavior triggered by a single-puffed pheromone stimulus

Therefore, with increasing frequency of stimulation as it nears an odor source, the path of a moth becomes straighter with repeated straight-line walking (surge). In contrast, if the frequency of stimulation decreases, the path becomes a complex combination of zigzagging and looping. Thus, male moths orient toward the odor source by repeating the set and reset of the programmed zigzag walking pattern depending on the spatial and temporal distribution status of odorant in the air. The significance of this behavioral pattern for pheromone source localization has been confirmed by mobile robots, with real antennae acting as pheromone sensors (Kuwana et al. 1999; Kanzaki et al. 2005) (Fig. 3). Silkmoths are not able to fly although they flap their wings during the pheromone tracking behavior. However, the overall mechanisms are regarded as the same as those in flying moths, because wing retraction which is a sign of steering flight in flying moths coincidences with the steering walking in silkmoths (Kanzaki 1998).
Fig. 3

Insect-behavior-based robot and a male silkmoth. Behavior-based model was applied to the robot as a controller. Moth antennae are used as pheromone sensors

Although pheromone (bombykol) alone is sufficient to localize the odor source, there are several factors that affect the locomotion pattern of the pheromone-searching behavior in the silkmoth. For example, as observed in other moth species (Ochieng et al. 2002), a mixture of the host plant odorant affects the sensitivity to the pheromone. In the silkmoth, a mixture of bombykol and cis-3-hexen-1-ol increases the sensitivity of males (Namiki et al. 2008). Moreover, serotonin levels in the brain also affect the sensitivity to pheromone. It is revealed that the serotonin level in the brain changes within a day and the dynamics are similar to that of pheromone sensitivity of male moths. This observation was verified by serotonin injection to the brain (Gatellier et al. 2004; Hill et al. 2002, 2003). In a related species, Manduca sexta, the effect of serotonin on individual neurons has been examined and it was revealed that serotonin application changes membrane properties of antennal lobe projection neurons and changes their excitability (Mercer et al. 1995, 1996; Kloppenburg et al. 1999).

Insect-controlled robot

In order to evaluate the behavioral capabilities of the insect, a mobile robot driven by an intact insect was fabricated. We call this insect–machine hybrid system ‘insect-controlled robot’ (Emoto et al. 2007). This insect-controlled robot is a two-wheeled robot equipped with behavioral measurement, signal processing and motor control subsystems. Details of this insect-controlled robot are reported by Emoto et al. (2007) and Ando et al. (2013).

The silkmoth is mounted on a polystyrene ball floated by air blown from the bottom of the robot. The ball is rotated according to the walking, and the rotation is sensed by an optical sensor set in a vertical position. The measurement subsystem uses the same mechanism as the system used in behavioral studies (Hedwig and Poulet 2004; Sakurai et al. 2011). An on-board microcontroller calculates the insect’s trajectory from the sensor output and controls the rotation of two DC motors on opposite sides. To supply the air containing odorant to the moth antenna on each side, two fans at the front of the robot are used. The airflows from these fans are separated by a separator to enhance the chemotaxis on the basis of bilateral olfactory cues (Fig. 4). The front to back airflow is necessary to enable the robot to localize the pheromone source at a success rate of 100 % in a wind tunnel, which is the same performance as that of walking silk moths in the same experimental condition. It also cancels the upward airflow from the bottom of the ball.
Fig. 4

Insect-controlled robot and a male silkmoth. a Insect-controlled robot. Dashed line indicates outline of a transparent separator for supplying airflow to each antenna. b A tethered moth on a ball (after Ando et al. 2013)

The insect-controlled robot enables us to alter the interaction between insect and environment quantitatively at any time by manipulating the motor characteristics of the robot. Focusing on visual information, it is possible to use virtual reality space connected with insect locomotion or physiological signals rather than mobile robots behaving in the real world (Gray et al. 2002; Butala et al. 2007; Takalo et al. 2012). In these systems, researchers are also able to manipulate interaction between insects and the environment via visual information. However, it is difficult to present olfactory information to insects in this virtual space. Using a mobile robot is the solution which enables us to stimulate on-board insects with odorant flowing in the real world and alter the interaction between insects and the environment.

A male silkmoth is used as a ‘driver’ and its behavioral responses to imposed perturbations during odor tracking are investigated. By employing this insect-controlled robot with the manipulation of the turning bias, we are able to put the moth into an extraordinary situation in which the moth is required to change its behavior, using appropriate sensory-motor systems for successful orientation. This application of perturbations to the sensory-motor system of the moth performing odor tracking will be an effective way to investigate the ability of the moth to compensate for the perturbation in sensory feedback (Ando et al. 2013).

With manipulation of the turning bias, the on-board moth shows a turning preference towards the non-biased direction (Fig. 5). When the robot is manipulated to induce turning bias, it locates the odor source by compensatory turning induced by the on-board moth. Shifting of the orientation paths to the odor plume boundaries and decreased orientation ability caused by covering the visual field (Fig. 5c) suggest that the moth steers with bilateral olfaction and vision to overcome the bias. At the boundaries of the plume, the moth can detect high gradient of the pheromone concentration by two antennae and steer into the plume (toward the side of higher concentration), which is facilitated by front fans (Fig. 4). These results indicate that steering based on the positive chemotaxis with bilateral olfactory input and optomotor response (visually guided compensatory movement against involuntary displacement from a straight course) is involved in the compensatory turning (Fig. 5d). Though the interaction between these two sensory-motor responses is still unknown, the similar conclusion is obtained in the study of phonotaxis and optomotor response in crickets, employing mobile robot (Webb et al. 2004). In the study, the authors demonstrate a capability of the multi-sensory integration model to compensate the imposed tuning bias and propose inhibitory interaction between phonotaxis and optomotor response for steering control.
Fig. 5

Manipulation of the insect-controlled robot. a Normal motor setting. Time course of turn angles between an on-board moth (magenta) and the robot (blue) during behavior corresponded with each other. b Turning bias towards clockwise (CW) direction. Time course of turn angles indicates that counterclockwise (CCW) turning of the moth stabilized turn angle of the robot. c Pheromone tracking trajectories of the robot in three conditions. Each plot indicates position of the on-board moth. The moth was surrounded by a transparent sheet in normal and turning bias conditions while by a white paper in turning bias + covered condition to suppress optic flow input. Pheromone receptive zone represents an area where the on-board moth can receive the pheromone (after Ando et al. 2013). d Mechanisms of the compensatory turning (blue arrows) against the turning bias (a red arrow)

Neural processing mechanisms underlying pheromone source searching behavior

As described above silkmoths are thought to be capable of generating adaptive behavior by modification and compensation of neural processing in the brain that drives pheromone source localization. The neural basis of pheromone source localization of the male silkmoth is summarized below. Detailed olfactory processing in moths is described by Haupt et al. (2009).

Silkmoths have a pair of the antennae. There are three main types of olfactory sensillae on the antennae (i.e., sensilla trichodea, sensilla basiconica, and sensilla coeloconica) (Koontz and Schneider 1987). The antenna has a total of 50,000 olfactory receptor neurons (ORNs) (Schneider and Kaissling 1957; Koontz and Schneider 1987). Each sensilla trichodea (~17,000) contains a pair of pheromone responding ORNs. One ORN expresses BmOR1 which selectively responds to bombykol, the major component of the pheromone and the other one expresses BmOR3 which selectively responds to bombykal, the minor component (Sakurai et al. 2004; Nakagawa et al. 2005). Each ORN expresses a single olfactory receptor protein. Axons of the ORNs project to the first order olfactory center, the antennal lobe (AL). Pheromonal information is processed by conspicuous sexually dimorphic structures in the silkmoth brain. The AL consists of a dorsally located male specific macroglomerular complex (MGC) and ventrally located ordinary glomeruli (OG). The MGC is further divided into compartments called the toroid, cumulus, and horseshoe (Kanzaki et al. 2003). Native projection patterns of pheromone receptor neurons were examined using male moths bearing EGFP driven by BmOR1 or BmOR3-GAL4/UAS system. Axons of BmOR1-expressing ORNs terminate exclusively in the toroid, while those of BmOR3-expressing ORNs project into the cumulus (Sakurai et al. 2011).

The AL contains two major types of neurons besides the projections of ORNs. Projection neurons (PNs) are principal neurons which transmit olfactory information from the AL to higher olfactory centers, the lateral protocerebrum (LPC) and the calyx of the mushroom body. Local interneurons are intrinsic cells which connect individual glomeruli, most of which are GABAergic (Waldrop et al. 1987; Christensen et al. 1993, 1998; Iwano and Kanzaki 2005; Seki and Kanzaki 2008). PNs innervating the toroid and cumulus selectively respond to bombykol and bombykal, respectively. Thus, each compartment processes independent information (Kanzaki et al. 2003). Identification procedures for OGs have been established in the silkmoth, and there are approximately 60 ± 2 glomeruli in the AL (Kazawa et al. 2009).

The AL PNs project to a stereotypic locus in the LPC and the LPC has a map of olfactory input from the AL (Marin et al. 2002; Wong et al. 2002; Nishino et al. 2003; Tanaka et al. 2004; Seki et al. 2005). In the silkmoth, the axonal projections of PNs processing the major pheromone component, bombykol is different from the other PNs. Immunocytochemistry with anti-cGMP appears to stain the PNs that process bombykol information (Seki et al. 2005). Using the simultaneous staining of individual PNs and cGMP antibody staining, an olfactory map in the LPC has been revealed in the silkmoth. Toroid PNs selectively project to a triangle-like shape region, called the delta area of inferior lateral protocerebrum (ΔILPC). The minor pheromone component, bombykal is processed in the cumulus of the AL and then sent to the lateral part of the ΔILPC, partially overlapping with the area of axonal projections of toroid PNs. In contrast, OG PNs send axonal projections to the lateral portion of the LPC and there seems to be no overlap between projections of OG PNs and MGC PNs. Axonal projections of AL PNs seem to be divided into 3 areas, ΔILPC, lateral part of the LPC, and ventral part of the LPC (Fig. 6) (Seki et al. 2005; Namiki and Kanzaki 2011).
Fig. 6

Anatomy of neuropils in a silkmoth brain. a, b Frontal and dorsal view of silkmoth brain. Major neuropils are reconstructed. Neuropils related to pheromone information processing, macroglomerular complex (MGC) in the antennal lobe (AL), delta area of the inferior lateral protocerebrum (∆ILPC), superior medial protocerebrum (SMP) and lateral accessory lobe (LAL) are shown in color. A anterior, D dorsal, P posterior, V ventral. c Proposed major pheromone processing pathways in the brain

The axons of MGC PNs are confined to a relatively small area within the calyx of mushroom body. In contrast, the axons of OG PNs are more widely distributed (Seki et al. 2005; Namiki et al. 2013). PN axons for the minor pheromone component cover a larger area than those for the major pheromone component and partially overlap with those innervating nonpheromonal glomeruli (i.e., OGs), suggesting an integration of the minor pheromone component with plant odorants in the mushroom body (Namiki and Kanzaki 2011; Fukushima and Kanzaki 2009).

In order to search for the pheromone processing pathways in the rest of the brain, we focused on the superior medial protocerebrum (SMP), for two reasons, first because the SMP contains many sex pheromone responsive neurons in comparison with other regions. Second, we found pheromone responsive neurons that connect the SMP and the ΔILPC, and then the SMP and the lateral accessory lobe (LAL), the region where the steering (command) information is generated (Kanzaki et al. 1994; Mishima and Kanzaki 1999; Wada and Kanzaki 2005). Further, SMP neurons respond to visual stimulation, suggesting that the SMP might integrate multimodal sensory information. We speculate pheromonal information is transferred from the ΔILPC to the LAL through the SMP (Fig. 6).

The LAL is a neuropil located in the ventral portion of the protocerebrum (Fig. 6). Based on the fact that descending neurons which have dendritic arborization in the LAL and the adjacent ventral protocerebrum (VPC), link with thoracic ganglia, and show pheromone responses, the LAL and VPC could be important olfactory neuropils in which the output of the brain in response to olfactory stimuli is generated (Kanzaki et al. 1994). The LAL/VPC consists of 2 types of descending neurons called Group I and Group II, all of which are morphologically and physiologically identified (Fig. 7) (Mishima and Kanzaki 1999; Wada and Kanzaki 2005). The LAL is further divided into two regions—the upper and lower divisions. The adjacent VPC is divided into three compartments. Besides the descending neurons, the LAL/VPC consists of two types of neurons—bilateral neurons, which link the LAL/VPC in both hemispheres, and unilateral neurons whose branching is unilaterally restricted to the LAL/VPC (Iwano et al. 2010).
Fig. 7

Reconstructed confocal images of all types of group I (GI) and group II (GII) descending neurons (DNs). Each type of DNs was intracellularly stained with Lucifer yellow. Posterior view. Inset schematic representation showing the structure of GI and GII DN in the whole brain. Scale bars 100 μm. AL antennal lobe, AN antennal nerve, D dorsal, LAL lateral accessory lobe, Oe oesophageal foramen, OL optic lobe, V ventral, VNC ventral nerve cord (after Wada and Kanzaki 2005)

Specific subsets of descending interneurons (Group IIA and IID; Fig. 7) (Mishima and Kanzaki 1999; Wada and Kanzaki 2005) show a characteristic state-dependent activity resembling an electronic flip-flop that has two distinct firing frequencies: high and low (Fig. 8) (Olberg 1983; Kanzaki et al. 1994). Switching back and forth between the two states occurs upon pheromonal stimulation. Such flip-flops are the fundamental element of ‘memory’ in electronic circuits. When flip-flop activity (FF) is recorded simultaneously from the right and left connectives, they show an antiphasic relationship (Fig. 8) (Kanzaki et al. 1994).
Fig. 8

Physiology of the cervical nerve 1 neck motor neurons (cv1 NMNs). a Activity pattern of the right and left cv-1 NMNs. Activity patterns correlate between a right (R cv1) and a left (L cv1) cv1 NMNs. b State transition in cv1 NMNs. State transition from high state to low state and from low state to high state. Overlay plots of the firing frequency (0.2 s bins) in multiple trials. c State transition in flip-flop descending neurons. State transition from high state to low state and from low state to high state. Overlay plots of the firing frequency in multiple trials. Gray band indicates onset of pheromonal stimulation and approximate period of state transition of cv-1 NMN (after Mishima and Kanzaki 1998)

These flip-flopping descending neurons seem to have synaptic contacts with a neck motor neuron (i.e., cv1 NMN; Figs. 8, 9) which regulates horizontal sideways movement of the head (Wada and Kanzaki 2005). This head movement is accompanied by the zigzag walking and the timing to alter the walking direction and horizontal sideways movement of the head is synchronized during pheromone-mediated behavior (Kanzaki and Mishima 1996). Head movement is controlled by the action of 11 pairs of neck muscles. Horizontal sideways movement of the head is caused by the action of the following three neck muscles: the ventral muscles (v) and the cervical ventral muscles 1 and 2 (cv1, cv2) (Fig. 9) (Mishima and Kanzaki 1998). These muscles are innervated by neck motor neurons passing through the second cervical nerve b (2nd CNb). Therefore, the activity of NMNs contained in the 2nd CNb which cause horizontal sideways movement of the head is a reference of pheromone-triggered zigzagging behavior (Fig. 9) (Kanzaki and Mishima 1996; Mishima and Kanzaki 1998; Wada and Kanzaki 2005).
Fig. 9

Anatomy of neck motor system. a Structure of the neck cuticle and candidate neck muscles which control sideways head movement. View from diagonally behind and ventral view are shown. c cerviale, f furca, poc postocciput. b Schematic drawing of nerve innervation to neck muscles. Dorsal view. cv cervical ventral muscle, d dorsal muscle, t tentorial muscle, v ventral muscle, 1st CN first cervical nerve, 2nd CN second cervical nerve, IN prothoracic ganglion nerve. c Confocal images of a lucifer yellow-stained neck motor neuron in the cv1 nerve. Frontal view. Inset schematic representation of the cv1-NMN. d Reconstructed confocal images of a double-labeling of neurobiotin-stained GII-A (red) and Lucifer yellow-stained cv1-NMN (green) (Left), and GII-D (red) and cv1-NMN (green) (Right). Posterior view (after Mishima and Kanzaki 1998; Wada and Kanzaki 2005)

Another subset of descending neurons with dendritic arborizations in the LAL/VPC has also been found to have synaptic contact with the NMN (i.e., cv1 NMN). These are Group IIC descending neurons which respond with brief excitation (BE) to pheromonal stimulation (Fig. 7) (Wada and Kanzaki 2005). Moreover, as shown in Fig. 8b and c, the NMN responds with BE and FF to pulsed pheromonal stimulation of the antenna (Mishima and Kanzaki 1999; Wada and Kanzaki 2005). Since the activity pattern of the NMN is a reference of pheromone-triggered zigzagging behavior, these results indicate that pheromone-mediated programmed orientation behavior is initiated by distinct groups of descending neurons. It was proposed that straight-line walking is controlled by the BE of Group IIC descending neurons, while the subsequent zigzagging turns and looping are controlled by the FF of Group IIA and IID descending neurons (Fig. 7) (Mishima and Kanzaki 1999; Wada and Kanzaki 2005). Group IIA and Group IID descending neurons might connect the several groups of neurons in the central pattern generator in the thoracic motor center to initiate the zigzag walking.

Brain–machine hybrid system (BMHS)

To understand the neural basis of adaptive behavior, recording dynamic activities of a brain during adaptive behavior is required. In a previous study, Melano (2011) suggested a possibility to drive a robot by neural signals recorded from an insect. We have proposed a novel experimental system that we call “brain–machine hybrid system (BMHS)”. The BMHS is a two-wheeled mobile robot (e-puck, EPFL, Switzerland) driven by steering (command) signals recorded from the brain of the silkmoth (Fig. 10).
Fig. 10

Brain–machine hybrid system. a Picture of the brain–machine hybrid system. b Expanded picture of a silkmoth mounted on the system. cBlock diagram of the brain–machine hybrid system (after Minegishi et al. 2012)

Putative steering (command) signals corresponding to walking of the silkmoth were selected. As described above, the activity of the neck motor neurons (NMNs) contained in the 2nd CNb, which cause horizontal sideways movement of the head, is a reference of pheromone-triggered zigzagging behavior. So the activities of the 2nd CNb were used as (command) signals for steering the robot.

It was expected that the silkmoth is controlled by two different steering (command) signals (i.e., BE and FF) as suggested above, the activity of the 2nd CNb would drive the robot toward the odor source. Further, as in the insect-controlled robot, the BMHS also had to modify its behavior using the changing sensory-motor feedback (Fig. 5).

Activity of neck motor neurons for steering (command) signals of the robot

Neck motor systems of insects have been well-studied among various species (fly: Strausfeld et al. 1987; Milde et al. 1987; Gilbert et al. 1995; Gronenberg and Strausfeld 1991; Gronenberg et al. 1995; Parsons et al. 2006; Huston and Krapp 2008; Wertz et al. 2012; honeybee: Schröter et al. 2007; Berry and Ibbotson 2010; Hung et al. 2011; locust: Shepheard 1973, 1974; moth: Eaton 1974; Dombrowski et al. 1990; Kanzaki and Mishima 1996; Mishima and Kanzaki 1998). As described above the activity of neck motor neurons (NMNs) contained in the 2nd CNb which cause horizontal sideways movement of the head is a reference of pheromone-triggered zigzagging behavior in the silkmoth (Fig. 8) (Kanzaki and Mishima 1996; Mishima and Kanzaki 1998; Wada and Kanzaki 2005). As the 2nd CNb contains five motor neurons (Mishima and Kanzaki 1998), physiological properties of these five motor neurons have been characterized. Neural activities of left and right 2nd CNbs were recorded bilaterally using glass suction electrodes. Figure 11 shows neural activities of the 2nd CNbs. Judging from the spike amplitude, five units are clearly recorded.
Fig. 11

Responses of 2nd CNb to optic flow stimuli. a Left and right 2nd CNb responded to leftward and rightward optic flow, respectively. b Selection of steering information from signals: we sorted spikes into 5 units, and numbered units according to spike amplitude from the smallest to the largest. cHistogram shows spiking of each numbered unit (bin is 0.1 s) (after Minegishi et al. 2012)

Since the 2nd CNb is also activated by optical flow which induces an optomotor response, physiological properties of each unit in response to optical flow stimulation were characterized. As in Fig. 11, the activity of each unit is shown as spike rate histograms with bins of 0.1 s. In response to optical flow stimulation, 4 out of 5 units showed consistent direction sensitive responses. In contrast, remaining fifth unit did not respond to the optical flow stimulation but constantly showed tonic activity (Fig. 11). As spike amplitude of the fifth unit was significantly smaller than other four units, these four units were sorted by spike amplitude and the total spiking activity of these four units was used for steering (command) signals of the robot.

A moth was set ventral-side up on the robot with its abdomen, and all legs and dorsal part of the thorax were removed. As shown in Fig. 10b neural activities of the 2nd CNbs were recorded bilaterally using glass suction electrodes (Minegishi et al. 2012). Silkmoths displayed normal behavioral patterns and orientation behavior when upside-down on the ceiling of a wind tunnel (personal observation).

Spike-behavior conversion rule

The spiking activities of the 2nd CNb were converted to control the robot movements as shown below. Since behavioral experiments indicate that the angular velocity of the body axis is proportional to the head angle (Minegishi et al. 2012), this correlation was used to develop a spike-behavior conversion rule to control the robot. The movement of a two-wheeled moth-sized mobile robot was assumed in a two-dimensional coordinate system. Forward velocities of the left and right wheels (i.e., \(v_{\text{left}}\) and \(v_{\text{right}}\)) were calculated proportionally to the right and left spike rate per 0.1 s (i.e., \(N_{{{\text{right}}\_{\text{spike}}}}\) and \(N_{{{\text{left}}\_{\text{spike}}}}\)) (Eqs. 1, 2). Depending on this sampling rate (10 Hz), the robot was also controlled at 10 Hz.

The proportional constant was given by dividing the average walking velocity of a moth \(v_{\text{ave}}\) (i.e., 2.6 mm/0.1 s) by the average spike rate of NMNs (i.e., \(N_{\text{ave}}\) = 8 spikes/0.1 s). These parameters were acquired from the average walking speed and spiking rate of moths under conditions of 2 Hz of pheromonal stimulation in a wind tunnel.

The moth-sized robot had a width \({\text{w}}\) and the center of the robot has forward velocity \(v_{\text{O}}\) (Eq. 3). Angular velocity \(\dot{\theta }\) of this robot was also calculated from Eqs. 1 and 2.
$$v_{\text{right}} = \frac{{N_{{{\text{left}}\_{\text{spike}}}} }}{{N_{\text{ave}} }}v_{\text{ave}} $$
$$v_{\text{left}} = \frac{{N_{{{\text{right}}\_{\text{spike}}}} }}{{N_{\text{ave}} }}v_{\text{ave}} $$
$$v_{\text{o}} = \frac{{v_{\text{right}} + v_{\text{left}} }}{2} $$
$$\dot{\theta } = \frac{{v_{\text{right}} - v_{\text{left}} }}{{2{\text{w}}}}. $$
Then, the movement of the left and right wheels of the hybrid system used in this implementation (width \({\text{W}}\)) (i.e., \(v_{\text{LEFT}}\), \(v_{\text{RIGHT}}\)) was calculated when it moved to the same location as the moth-sized robot described below (Eqs. 5, 6). The center of the hybrid system was calculated to that of a moth independent of its size. The actual widths of a moth and the hybrid system were 10 and 115 mm, respectively.
$$v_{\text{LEFT}} = \frac{{v_{\text{right}} + v_{\text{left}} }}{2} - \frac{\text{W}}{\text{w}}\frac{{v_{\text{right}} - v_{\text{left}} }}{2} $$
$$v_{\text{RIGHT}} = \frac{{v_{\text{right}} + v_{\text{left}} }}{2} + \frac{\text{W}}{\text{w}}\frac{{v_{\text{right}} - v_{\text{left}} }}{2}. $$

Odor source searching behavior by BMHS

The behavioral pattern and odor source orientation behavior of the BMHS were investigated in a wind tunnel. First, it was tested whether the BMHS could show the characteristic behavioral pattern with zigzagging manner like an intact moth. The BMHS was set downwind from the pheromone releasing site, and walking trajectories induced by a single puff of pheromone stimulation to the antennae were recorded. Elements of the programmed behavioral patterns (i.e., surge, zigzagging turns and loop; Fig. 2) were consistently observed (Fig. 12) (Minegishi et al. 2012). This indicated that the 2nd CNbs’ activities were adequate signals to be converted into the odor source searching behavior of the silkmoth. Then, it was investigated whether the BMHS could orient toward a pheromone source in the wind tunnel. The BMHS oriented toward the source with more than 60 % success rate (Minegishi et al. 2012). In order to evaluate the performance of the BMHS, the trajectories of the moths with those of the BMHS toward the pheromone source were compared. Heading angles of the moth and the BMHS were calculated with respect to the upwind direction every 0.1 s (Fig. 13). As shown in Fig. 13, both showed similar tendencies in heading angles in response to a range of stimulation frequencies (Minegishi et al. 2012).
Fig. 12

Programmed behavioral pattern of a silkmoth performed on the brain–machine hybrid system. We plotted angle of the hybrid system body axis. The moth on the hybrid system received a pheromone stimulus (bombykol; 100 ng, 500 ms) at 0 s on this graph (filled triangle), and showed the subsequent programmed behavioral pattern (after Minegishi et al. 2012)
Fig. 13

Heading angle of a silkmoth (bars) and the hybrid system (dots on the bars) in response to a range of frequencies of pheromonal stimulation. Heading angle with respect to the upwind (θ) is shown as a cos θ as directional index (after Minegishi et al. 2012)

As described above, if the silkmoth is controlled by two different command signals (i.e., BE and FF), the activity of the NMNs will drive the robot toward the odor source. The significance of these two signals as steering (command) information was verified for the first time under real circumstances using this BMHS.

Further, the turning bias was applied to the BMHS as applied to the insect-controlled robot (Fig. 5). The BMHS was put into different-gain motor conditions in which the moth is required to modify its neural activities using its sensory-motor system for successful orientation. It was examined whether the activities of the 2nd CNbs were modified by externally given unintentional movement. The activity of the 2nd CNb of the non-biased side was more highly activated than the biased side. The spiking activity of the biased side was rather reduced (Minegishi in preparation).

Thus, using the BMHS it is possible to record moment to moment neural activities for the first time from the brain during unintentional conditions by turning bias under real circumstances. Although BMHS research is still ongoing, this system will contribute to evaluation of the adaptability of behavior, and will increase understanding of dynamic neural networks that generate adaptive behavior.

Conclusions and future work

The neural mechanism for pheromone source localization in the male moth is one of the most tractable pathways for understanding how animals control behavior, despite environmental perturbations (Haupt et al. 2009). Dynamical neural mechanisms generating the steering control outputs relayed by descending neurons in the silkmoth is currently our major challenge. The first implementations have started to appear in the area of odor source localization as ‘insect–machine hybrid system’, coupling biological information processing to artificial effectors (Emoto et al. 2007; Minegishi et al. 2012; Ando et al. 2013). In the insect–machine hybrid system the mobile robot is controlled by an insect through its behavior or neural activity recorded from the brain. This system will contribute to evaluation of the adaptability of behavior, and to a better understanding of the dynamics of neural networks.

Moreover, the relatively small size of moth brains and a large body of identified neuron data are now being used to attempt rebuilding behaviorally relevant circuits of the moth brain by means of realistic biophysical simulations. We are now constructing a large-scale realistic neural network model with multicompartments on the K-computer (world’s third fastest supercomputer) (Ikeno et al. 2012).

Genetic engineering also plays an important role for brain science. Genetic manipulations have become feasible in the silkmoth (Yamagata et al. 2008; Sakurai et al. 2011). Recently we have expressed a light-gated cation channel, channelrhodopsin-2 (ChR2), on pheromone-sensitive olfactory receptor neurons (ORNs) and some brain neurons using GAL4/UAS system (Tabuchi et al. submitted). This makes it possible to control spiking activity with millisecond resolution in ChR2 expressed cells. We have succeeded to initiate the normal zigzagging behavior in response to light stimulation in the ChR2 expressed moth on the ORNs. Optogenetics will lead a new phase in understanding olfactory processing in the brain. Thus, a wide variety of methodological approaches are available in the silkmoth.

Combining these approaches will enable us to use the full potential of the features of insect brains as model systems for understanding the dynamical neural substrates of adaptive behaviors for the first time. As well as being of biological interest, this topic is also of interest for robotics, because robots also need to execute tasks in changing environmental conditions. The brain model is expected to control mobile robots adaptively and the bottom up study on the basis of insect behavioral neuroscience is fundamental approach to realize the biomimetic odor tracking. The insect–machine hybrid system reviewed here will also be helpful for evaluation of the brain model when it is implemented into the robots, since the hybrid robot is a blueprint of the biomimetic study.

Copyright information

© Springer-Verlag Berlin Heidelberg 2013