The silk moths used in the experiments were purchased in the pupal stage (Kinshu × Showa; Ehine Sansyu Co. Ltd., Ehime, Japan). Emerged individuals were kept in an incubator at 16 ° C and allowed to acclimate to the laboratory room temperature (24- 26 ° C) for at least 15 min before the experiment. Silk moths used for behavioral experiments and EAG recording were 2–5 days old after eclosion. When attaching markers to the silk moth, an adhesive (G17, KONISHI Co.,Ltd., Osaka, Japan) that does not affect behavior was used (Shigaki et al., 2016).

The robot system was built based on a commercially available laser processing machine (FABOOL Laser Mini 150 cm × 150 cm, SMART DIYs, Yamanashi, Japan). Details will be given in the following section, and the robot system is equipped with a high-speed camera (DFK33UX273, The Imaging Source, Bremen, Germany), an amplifier (ISD2PAD, Oisaka Electronic Equipment Ltd., Japan) for measuring physiological responses, and a small microcomputer for log transfer. Operations from acquiring camera images to controlling the motor of the laser processing machine are carried out at 100 Hz. The fixtures for attaching the camera and amplifier to the robot system were created using a 3D printer (ABS; Creator3 Pro, Flashforge, Zhejiang, China) and laser processing machine (Etcher Laser Pro, SMART DIYs, Yamanashi, Japan).

It must be recorded through an appropriate amplifier because EAG is a small electrical signal. Based on previous research, the EAG amplifier is equipped with high-pass (0.1 Hz) and low-pass (300 Hz) filters and has an amplification rate of approximately 1,000 times (Shigaki et al., 2024). Electrodes (L-shaped pin headers, Useconn Electronics Ltd., Zhejiang, China) for attaching an antenna are attached to the amplifier, and conductive gel (Spectra360, Parker Laboratories, United States) is used to promote the connection between the antenna and the electrodes. EAG was recorded at a sampling rate of 1 kHz. The acquired EAG data was passed through a software low-pass filter with a cutoff frequency of 10 Hz.

R software (R Core Team) was used to create the histograms. In addition, video footage was analyzed using DeepLabCut (DLC) (Mathis et al., 2018). After analysis with DLC, trajectories and EAG data were integrated using a self-coded MATLAB code (2023b, MathWorks, MA, United States). Information entropy was calculated using Python with a self-coded program.

The proposed system used a high-speed camera (DFK33UX273, The Imaging Source, Bremen, Germany) to measure the moving direction and speed of the insect and tracked the movement of the silk moth without contact. Images were acquired at 100 fps with a resolution of 1280 × 720 . The acquired images were processed using the open-source library OpenCV (ver. 3.4.7, Intel, CA, United States). The maximum translational and angular velocities of the silk moth have been reported to be 32.8 mm/s and 1.0 rad/s, respectively (Takashima, 2010). By using a high-speed camera with 100 fps (Hz) to keep the translational error below 5 mm, the expected error is 0.33 mm/frame, suggesting that the control objective can be theoretically achieved. Therefore, we expected that visual feedback could be used to track the insect sufficiently.

However, to set the control cycle to 100 Hz, the process from measurement to calculation of the control command must be completed within 10 m. Consequently, we need to minimize the load on the image processing system. As in a previous study (Shigaki et al., 2016), we attached two-color markers with diameters of 2 mm to the silk moth and implemented an algorithm to calculate the moving speed and direction of the insect using simple color extraction. In particular, the color of the markers was extracted in the HSV color space, and the center of gravity of each marker was calculated as the marker coordinate. The red marker placed near the head was defined as the position of the moth in the observation coordinate system, and the approximate line connecting the two markers was defined as the heading angle of the moth. The markers were treated as points after image processing. As demonstrated in a previous study (Shigaki et al., 2016), we confirmed that attaching these markers did not interfere with the moth movement.

This robot system was based on a laser processing machine (FABOOL Laser Mini 150 cm × 150 cm, SMART DIYs, Yamanashi, Japan). Stepping motors were attached to the X- and Y-axes, and the position of the end effector can be controlled by providing commands to the stepping motors. The position of the end effector can be controlled by transmitting the relative position information in G-code. Relative position was defined as the position error between the origin of the camera coordinate system ( Σ c ) calculated using the observation system and the insect, as shown in Figure 2A. This position error has the same meaning as the tracking error for evaluating tracking performance.

We employed a proportional-integral-derived (PID) controller to control the robot system. The Ziegler-Nichols’ closed-loop method was used to determine each parameter of the PID control, and the final fine parameter adjustment was performed empirically. The results of calculating the frequency response of the robot system using PID control are shown in Figure 2B. Consequently, the bandwidth was found to be approximately 2 Hz for both the x- and y-axes. The robot system has been suggested to be capable of tracking the silk moth while searching for an odor source because the period during which the silk moth changes its behavior significantly from side to side is approximately 1 Hz or less.

In addition, in a preliminary experiment, we measured the tracking error while the silk moth searched for an odor source. When the tracking error was plotted as a histogram, as shown in Figure 2C, we found that the error while tracking the silk moth was less than 5 mm on both axes, indicating that the control target was achieved.

An instrumentation amplifier (ISD2PAD, Oisaka Electronic Equipment Ltd., Japan) was attached to the end effector to measure electrophysiological responses. This study aimed to measure the odor reception state during an odor source search using an EAG; therefore, the instrumentation amplifier was customized for EAG measurement. The instrumentation amplifier was equipped with a 0.1 Hz high-pass filter and a 3 kHz low-pass filter and had an amplification rate of approximately 1000 times. The data recorded using this amplifier were collected on a small microcomputer (TinyPICO, Unexpected Maker, Melbourne, VIC, Australia) using an AD converter (MCP3208, Microchip Technology, AZ, United States) and then wirelessly transferred to a recording PC. The sampling rate and EAG recording cycle were set to 1 kHz.

The response profile obtained from EAG measurements in a static condition, where the distance between the odor source and the antennae is fixed, is illustrated in Figure 2D. As depicted in Figure 2D, the time taken to reach the peak following odor stimulation is defined as the response time ( T r ) , while the time required for the response to return from the peak to the baseline is defined as the recovery time ( T d ) (Shigaki et al., 2019a). In a static context, for single odor stimuli, the response time is typically within 10 milliseconds, and the recovery time generally exceeds 100 milliseconds. However, this is an example measured under static conditions, and these profiles may change if the odor is received in a dynamic situation. In addition, it is important to note that different receptor cells respond to different types of odors; therefore, EAG waveforms differ depending on the odorant. Moreover, in response to periodic odor stimuli, it shows attenuation characteristics as shown in Figure 2E. Figure 2F shows the results of the EAG measurement performed with the robot system turned on and the motor vibrating slightly. An odor stimulus of 1 Hz was emitted from an odor source located approximately 10 cm away. As shown in Figure 2F, the EAG response was measured in response to odor stimulus, demonstrating that the EAG can be measured even in the presence of electrical or mechanical noise.

EAG was used to estimate the level of odor reception by silk moths during odor source localization. In this study, the EAG measurements were performed using the antenna of another silk moth, i.e., the “third antenna,” as in the previous study (Vickers and Baker, 1994). The antenna used for the EAG was an adult male silk moth 2–7 days after emergence. The silk moths were purchased in the pupal state (Kinshu × Showa; Ehine Sansyu Co. Ltd., Ehime, Japan). The emerged silkworm moths were managed under a 16:8 h light:dark photoperiod at 18 ° C and 50–60 % relative humidity using an incubator (CN-40A, Mitsubishi Electric Engineering Co.).

The antenna of a silk moth was cut from its head and attached to the electrodes on an amplifier using a conductive gel (Spectra360, Parker Laboratories, NJ, United States). We confirmed in advance whether the EAG could be correctly recorded from the antenna attached to the electrodes by providing a single shot of the sex pheromone (Bombykol). The antenna was then fixed in the designated position on the end effector. A previous study examining the sensitivity and stability of EAG recording after antennae isolation has shown that the EAG response does not deteriorate within an hour after isolation (Martinez et al., 2014). In the silk moth EAG measurements, preliminary experiments similarly showed that there was no difference in response performance within about 45 min after isolation; therefore, we set the time from antenna isolation to the completion of the simultaneous EAG-behavior experiment to be within 30 min.

As illustrated in Figure 3A, odor source localization experiments were conducted in a flat open area. Bombykol, the sex pheromone of a female silk moth, was used as the odor source at the origin of the experimental field ( x , y ) = (0, 0)[m]; it was placed as the odor source. A 1000-ng concentration of bombykol was applied to a filter paper with a diameter of 10 mm. The pheromone-coated filter paper was replaced after each experiment. Air was blown over the filter paper at a flow rate of 4.0 L/min to disperse the odor into the surrounding space. The pheromone concentration employed in this experiment corresponds to the average amount of bombykol emitted by female silk moth (Fujiwara et al., 2014). Furthermore, females have been reported to release bombykol at a frequency of 0.79 ± 0.05 Hz (Fujiwara et al., 2014), and accordingly, in past behavioral experiments, a similar value was used, 1 Hz (duration: 0.2 s, interval: 0.8 s) (Yamada et al., 2021; Kanzaki et al., 1992). Therefore, in this study, the air used to deliver the odor was regulated by an electromagnetic valve, and the odor was emitted at a frequency of 1 Hz (duration: 0.2 s, interval: 0.8 s). A electric fan (YLS-18, Yamazen Corporation, Osaka, Japan) was positioned 0.5 m behind the odor source to facilitate the diffusion of the pheromone. The voltage was controlled so that the wind speed generated by the fan was approximately 0.6 m/s at the odor source. This wind speed was the same as in previous silk moth behavior experiments (Kanzaki et al., 1992). The silk moth was positioned at ( x , y ) = (150, 0)[mm], and the odor-source localization experiment commenced. The initial position from the odor source was also designed based on previous behavioral experiments with silk moths (Kanzaki et al., 1992). Localization was considered successful when the silk moth reached a position within a 10-mm radius of the odor source. However, localization was considered to have failed if the silk moth failed to achieve localization within 180 s. The movements of the silk moth were captured using a camera (FDR-AX45A, Sony, Tokyo, Japan) installed on the side of the field and converted into trajectory data. These experimental conditions were set to closely follow those used in previous experiments with freely walking silk moths (Kanzaki et al., 1992). The silk moths used in the odor source localization experiments were 2–7 days old after emergence. Before starting the behavioral experiment, the silk moths were removed from the incubator, markers were attached, and they were allowed approximately 1 h to become accustomed to the temperature of the behavioral experiment space. Fourteen silk moths were used in the odor source localization experiment.

The results of the odor-source search experiment revealed that all 14 individuals successfully localized the odor source (see Supplementary Video S1). The time required for odor-source localization was 24.8 ± 16.2 s. Due to interindividual variation in body size, discrepancies in locomotion speed were observed, resulting in variability in the time required for odor source localization. Nevertheless, it was demonstrated that all silk moths, regardless of differences in body size and associated locomotor speed, were capable of successfully localizing toward the odor source.

Figure 3B shows typical time-series data of EAG responses from a simultaneous EAG measurement experiment and behavior during odor-source localization. The time-series data revealed that odor reception began immediately after the experiment began. The EAG amplitude was defined as the EAG intensity, and the localization trajectory was displayed in a color corresponding to the EAG intensity to visualize the relationship between odor reception and behavior. Figure 3C shows the relationship between the EAG intensity and trajectory. The EAG intensity was normalized between 0 and 1; the closer the intensity to 1, the higher the concentration of the odor received. Places with strong EAG intensity are marked with numbers, as shown in Figures 3B,C. We found that strong odor reception resulted in straight-line movement. To quantitatively assess this phenomenon, we plotted the relationship between the EAG amplitude and translational speed, as illustrated in Figure 4A. The results indicate that once the EAG amplitude exceeds 1 mV, the translational speed becomes markedly greater compared to conditions in which the amplitude remains lower. This phenomenon is similar to previous research results that investigated the relationship between odor reception and straight-line behavior using a tethered measurement system (Takasaki et al., 2012). Experiments with this robot system have revealed that the same phenomenon is observed during actual odor-source localization. Furthermore, as shown in Figure 3C, the silk moth approached the odor source by exhibiting cast-like behavior commonly observed in flying insects, with 7 out of 14 individuals demonstrating this phenomenon. To quantitatively show which direction silk moths that exhibited cast-like behavior moved after encountering an odor, we calculated the moving ratio in the upwind, downwind, or crosswind direction as shown in Figure 4B. As a result, we found that while a silk moth moved in the upwind direction about 50 % , it also moved in the downwind and crosswind directions. This is because the behavioral strategy of silk moths differs from that of flying insects, and when they encounter an odor, they adopt a surge behavior in which they move in the encountered direction (Takasaki et al., 2012). However, looking at the global trajectories, it was suggested that they also move in the crosswind direction to gain a chance to encounter an odor.

The relationship between this trajectory and the EAG intensity for all individuals is depicted in Figure 3D as a three-dimensional color map. For comparison, an odor concentration map of the experimental field, generated using an artificial olfactory sensor (MiCS-5524 Gas Sensor Breakout, Adafruit Industries, NY, United States), is presented in Figure 3E. When employing the artificial sensor, it is evident that odor concentration increases in proximity to the odor source. However, from Figure 3D, we can see that for all insects, the EAG intensity tends to be higher during the early stages of searching, away from the odor source. A new insight emerged for comprehensive decision-making based on the time-series data in Figure 3B and the three-dimensional color map in Figure 3D: The amplitude of the EAG did not necessarily increase as the insect approached the odor source, where the odor concentration was expected to be higher. The ratio of the maximum EAG amplitude observed during localization behavior to the EAG amplitude measured closest to the odor source was calculated to be 0.47 ± 0.28 ( N = 14). This indicates that, in proximity to the odor source, the EAG output is reduced to approximately half of its maximal response. This suggests that the silk moth does not simply move toward areas with higher odor concentrations. As one approaches the odor source, exposure to odor stimuli becomes increasingly periodic; however, EAG responses exhibit an attenuation characteristic in response to such periodic stimulation. Figure 2E shows a typical EAG response to 0.7 Hz odor stimulation, with the amplitude decreasing after multiple odor stimulations. This decrease in amplitude recovered to some extent if the antenna did not receive any odor stimulation for approximately 5 s, suggesting that it reduced the response sensitivity to continuous odor stimuli near the odor source. Instead, it effectively reduces the distance to the odor source based on other cues. Next, we evaluated these experimental results using an information-theoretic approach to further analyze the behavior of the insects in greater detail.

The factors enabling odor source localization in the silk moth ( B o m b y x m o r i ) were analyzed from an information theory perspective. In particular, a quantitative analysis was performed by calculating the information entropy during the odor source localization behavior of the silk moth. The link between information entropy and odor source localization was first proposed using infotaxis, which does not rely on odor gradients to locate the odor source (Vergassola et al., 2007). Hernandez et al. (Hernandez-Reyes C. A. et al., 2021) reported the basic concept of using information entropy to analyze insect behavior. In this study, the analytical method of Hernandez et al. was applied to the results of the simultaneous EAG behavior measurement experiments. In this approach, the calculation is based on the timing of the insect’s encounter with the odor rather than on odor concentration. In the EAG data, the timing of the peak (falling timing) of the EAG response was automatically detected using the MATLAB peak detection function (findpeaks, 2023b, MathWorks, MA, United States). The peak was then used as the timing for encountering the odor.

Fundamentally, information entropy S t is calculated from a probability density function, as shown in Equation 1. S t = − ∫ P t ⁡ ln P t d r , (1) where, r represents the location of the searcher (silk moth). This probability density function is computed based on Bayes’ theorem, as expressed in Equation 2. The likelihood function used in the calculation of the probability density function is updated at each time step based on the encounter rate with the odor, as shown in Equation 3. P t r 0 = L Γ t | r 0 ∫ L Γ t | r x d r x (2) L ( Γ t | r 0 ) = e − ∑ ∫ V i R r t ′ | r 0 t ′ ∏ i = 1 n R r t i | r 0 (3) Here, the V i ’s is the time intervals of absence of detections, r 0 represents the location of the odor source, and r x represents the location of the insect in the x direction at time t . The expected rate of odor encounter, R ( r | r 0 ) , was calculated using Equation 4 based on an estimation of the plume parameters, such as the wind speed W , the emission rate of the source E and the insect position in x direction. R ( r | r 0 ) = a E | r − r 0 | e − | r − r 0 | λ e x − x 0 W 2 D λ = D τ 1 + W 2 τ / 4 D (4)

Essentially, the basic principle of infotaxis is that when the searcher is at P t ( r 0 | r ) at time t , the odor source is represented as a probability density function, in which the odor source is located at r 0 . Herein, the parameters used to calculate the information entropy are as follows: agent sensor size a = 10 mm, emission rate E = 1 , particle lifetime τ = 6.3 , and particle difficulty D = 0.057 . These values were obtained from Ref. (Vergassola et al., 2007; Hernandez-Reyes C. A. et al., 2021). Relative entropy is the ratio of the entropy at a certain time step to the entropy at the beginning of the search, i.e., S t / S 0 . Figure 5A shows the relative entropy trends of all insects. The gray line in Figure 5A represents the relative entropy of each trial, the red line represents the mean, and the thin red area represents the standard deviation. After the search began, we found that the insects all migrated with reduced information entropy. This suggests that silk moths may approach the target odor source by using the timing of the odor encounter as an indicator.

Estimating the probability of the odor source (i.e., the estimated position of the odor source) is possible while calculating the information entropy. Hence, we visualized the positions at which the silk moth estimated the odor source location during localization. Notably, it is unlikely that a similar mapping of the estimated location of the odor source occurs in the brain of the silk moth, which has a smaller number of neurons, and that the visualization of the estimated location in this study was simply introduced as an analytical method. Figure 5B shows a graph showing the time series changes in the error between the true odor source and the estimated source location for 14 silk moths. The gray line in Figure 5B represents the relative entropy of each trial, the red line represents the mean, and the thin red area represents the standard deviation. On average, the error tended to be smaller 5 s before localization. Figure 5C and (D) show the results of calculating the estimated odor source location (see Supplementary Video S2). The time series changes in information entropy in Figure 5C are related to the estimation results in Figure 5D, and the numbers on the triangles in Figure 5C correspond to the results in Figure 5D. Moreover, this trajectory follows the same data as in Figure 3C.

The odor source location was rather diffused in the early stages of localization when no information had been accumulated. However, as localization progressed and more odor information was gathered, the estimated odor source position moved closer to the actual odor source location. Finally, the odor source was estimated to be close to its actual position, suggesting that even the silk moth, which typically reacts reflexively to odor stimuli, possesses a strategy for efficiently approaching the odor source.

Thus, we suggest that selecting behavior appropriately based on the timing of odor encounters rather than simply moving in the direction of the strongest odor is adaptive for effective odor source localization. This study contributes to the understanding of the excellent navigation capabilities of insects by developing an insect-tracking robot system, enabling the measurement of the relationship between odor reception and behavior during the odor source localization of the silk moth, and analyzing the interrelation between them.