Search

Objective: Weakly electric fish prefer to hide in shelters in water in their natural habitat. If these shelters move, the fish follow these shelters by also using active sensing movements when necessary. Thanks to this unique feature, weakly electric fish are often used as animal models for active sensing research. However, a majority of the studies in the literature are limited to stationary environments unlike their natural habitats with continuous water flow. Different than the literature, this study aims to examine the effects of flow speed on the active sensing movements of weakly electric fish. Methods: In line with our aim, we built an experimental setup, where the fish can perform refuge tracking behavior under water flow. We connected the refuges a linear actuator that moves in a single linear dimension. The back-and-forth movements of the refuge generate the necessary visual and electrosensory cues for the fish. In our experiments, we used a complex stimulus as a combination of sinusoidal signals of 13 different frequencies in the range 0-2 Hz. We experimented with N=3 Apteronotus albifrons under various conditions, such as flow speed (0-20 cm/s), illumination (0-300 lux), and refuge structures (with/out windows). We performed five replicates for each fish under each sensory condition. Results: We verified that absence of illumination increased the amount of active sensing movements as previously reported in the literature. Besides, we observed that flow speed also increases the active sensing movements by reducing the tracking gain and increasing the phase lag. Conclusion: The modeling and transferring the contribution of active sensing movements to the engineering applications requires a solid understanding of its biological motivations. In this context, conducting these analysis in the presence of water flow is of critical importance in understanding the natural motivation of these fish. [ABSTRACT FROM AUTHOR]

Objective: The CNS use multisensory integration to reduce noise and uncertainty in sensory signals and better interpret its environment. This study investigates the mechanisms of multisensory integration in animals using probabilistic approaches. These probabilistic approaches have been derived from the assumption that our sensory perceptions are noisy and affected by uncertainty. Our studies aim to understand how the CNS solves the integration problem of this noisy and ambiguous sensory information. In this context, we develop models that describe how the brain combines signals from different sensory organs, using a dataset obtained with Eigenmannia virescens, a species of weakly electric fish. Methods: In this study, we investigate the superposition and Bayesian models on a dataset, where Eigenmannia virescens utilizes visual and electrosensory feedback signals to track a moving refuge. This dataset allows us to investigate how two sensory signals flowing in parallel are combined in the brain during multisensory behavioral control. The dataset includes refuge tracking experiments on five adult fish with ten replicates of six stimulus profiles. The dataset includes results with water conductivity of 150 and 500 μS/cm in dark. We investigate the dynamics of multisensory integration by analyzing the response of the fish to each stimuli using frequency response analysis. Results: In our study, analyses were carried out under the same environmental conditions at a specific stimulus frequency with three different experimental conditions: only visual, only electrosense, and both visual and electrosense applied simultaneously. We first tested a superposition model with fixed sensory weights. The unsuccessful results motivated us to use dynamically changing weights for the sensory information. Our findings showed that the fish dynamically modulates its sensory weights throughout the experiments. Conclusion: In this study, we tested a model that weights the sensory information inversely proportional to their variance, yielding the more consistent information to be weighted more. Our analysis showed that this model is not sufficient to explain the dynamic sensory reweighting process in these fish, since it did not bring a statistically significant improvement as compared to random distributions. We continue developing new models based on Bayesian filters for modeling the dynamic reweighting process during multisensory integration. [ABSTRACT FROM AUTHOR]

Objective: Animals perceive signals of different speeds and propagation patterns, such as light and sound, from their environments via various receptors in their bodies, and they construct a neural representation of their environment. Our goal is to identify the dynamics of the filtering mechanisms adopted by animals to combine signals perceived by different sensory organs during their free behaviors. Methods: In this study, we developed a unique experimental setup to analyze the multisensory integration of freely-swim-ming Danio rerio for target tracking during rheotaxis. We built a flow tunnel for zebrafish to perform rheotaxis. We placed a Dshaped cylindrical tube in the test area to obscure the flow in certain regions. Zebrafish naturally prefer to swim in the low flowspeed regions that occur behind these obstacles to avoid getting dragged with the flow. Zebrafish perceive the obstacles using the visual and mechanosensory stimulations in the water using their eyes and lateral line. A unique aspect of our design is that the visual and mechanosensory cues can be controlled independently, thanks to our special dual-motor actuation mechanism. This way, we can stimulate the relevant sensory organs synchronously or independently, enabling the identification of the dynamics of multisensory integration adopted by the CNS. Results: In this study, we repeated the target tracking experiments for N=3 zebrafish, and estimated the frequency response of the dynamics of the multisensory integration process. A key finding of our studies is that visual stimuli alone are not sufficient to initiate the fish, but mechanosensory stimuli are. Zebrafish reduces behavioral variability in the presence of both stimuli via multisensory integration. Conclusion: Our findings show that simple superposition models can not capture the dynamics of multisensory integration unless both sensory cues are available. In general, the dynamics of multisensory integration necessitate hybrid model structures. Superposition models are most likely to fit when both stimuli are available. [ABSTRACT FROM AUTHOR]

Objective: Zebrafish orient their bodies toward the flow to maintain their station. These fish prefer station-keeping behind the obstacles, which obscure the flow to minimize energy consumption. If these obstacles move, zebrafish track their movement to avoid getting dragged with the flow. We aim to analyze the movement mechanisms adopted by zebrafish during this target tracking behavior. Methods: We developed an experimental setup, which enables zebrafish to perform rheotaxis. We placed a D-shaped cylindrical obstacle with a diameter of 5 cm into the water to obscure the flow in desired regions. We move this obstacle via a linear actuator in the horizontal direction using sinusoidal trajectories of different frequencies within the range of 0.05-2 Hz. We conducted five repetitions for N=3 zebrafish. We recorded the response of the fish to the movements of the obstacle using a camera at 25-100 Hz. We used DeepLabCut to track the body movements of the zebrafish to understand the movement mechanisms adopted by zebrafish while tracking the target obstacles. We labeled 8-13 points on the body of the zebrafish and tracked these points on video records. We used these tracks to digitize the body oscillations during target tracking. Results: Zebrafish utilize tail oscillations to move their bodies in the lateral axis. Using these oscillations, zebrafish generate the necessary thrust to move their bodies to the left and right. We observed that these oscillations are correlated with the body movements of this fish. Conclusion: The correlation between the tail oscillations and the body movements of zebrafish allows us to infer a proxy of the neural controller using behavioral measurements. As an alternative to EEG and fMRI measurements, we plan to use tail oscillations as an approximation of the control output to understand the subblocks within the closed-loop feedback control system of zebrafish. [ABSTRACT FROM AUTHOR]

A common approach to understanding and controlling robotic legged locomotion is the construction and analysis of simplified mathematical models that capture essential features of locomotor behaviours. However, the representational power of such simple mathematical models is inevitably limited due to the non-linear and complex nature of biological locomotor systems. Attempting to identify and explicitly incorporate key non-linearities into the model is challenging, increases complexity, and decreases the analytic utility of the resulting models. In this paper, we adopt a data-driven approach, with the goal of furnishing an input–output representation of a locomotor system. Our method is based on approximating the hybrid dynamics of a legged locomotion model around its limit cycle as a Linear Time Periodic (LTP) system. Perturbing inputs to the locomotor system with small chirp signals yield the input–output data necessary for the application of LTP system identification techniques, allowing us to estimate harmonic transfer functions (HTFs) associated with the local LTP approximation to the system dynamics around the limit cycle. We compare actual system responses with responses predicted by the HTF, providing evidence that data-driven system identification methods can be used to construct models for locomotor behaviours. [ABSTRACT FROM AUTHOR]

Objective: Sensory salience defines how well a cue stands out from surroundings and plays a critical role in the behavioral control of animals. Our goal is to examine the effects of sensory salience on behavioral performance by modeling the target tracking behavior of zebrafish during rheotaxis. Methods: In this study, we developed a speed-controlled flow tunnel for zebrafish to perform rheotaxis. Zebrafish tend to maintain their position with minimal energy consumption by orienting their bodies downstream during rheotaxis. If an obstacle is placed in the test area, they interrupt the flow speed. We expect the fish to swim in these areas, where the flow speed is bearable. Our setup includes a linear actuator to control the movements of this obstacle. We generate visual and mechanosensory stimulus for the fish using semi-cylindrical (D prism) shaped obstacles placed in the water. In this study, we experimented with N=3 adult zebrafish with different flow speeds (~0.1 m/s - ~1 m/s), obstacle diameters (2, 3 and 4 cm), and presence/absence of illumination to change the sensory salience. We estimated the frequency response of the fish using different frequency stimulus profiles in the range of 0-2 Hz. Results: We observed that the fish performed better in the presence of light. This was expected as the fish can benefit from both visual and mechanosensory signals in this condition. Reducing the diameter of the D-prism-shaped obstacle also reduces the salience of both visual and mechanosensory stimuli. Accordingly, this change degrades the tracking performance of the fish. Conclusion: Zebrafish continuously integrate multisensory integration using visual and mechanosensory stimuli while target tracking during rheotaxis. Decreasing the sensory salience reduces the state estimation performance of the fish, and hence the target tracking performance. In the future, we aim to model the sensorimotor control process of zebrafish using sensory salience. [ABSTRACT FROM AUTHOR]

Objective: Animals benefit from diverse sensory organs that capture various sensory signals to make sense of the environment. These signals, combined by the central nervous system (CNS), serve to increase the quality and speed of processing for the sensory information. Our goal is to reveal how the CNS weights sensory signals from different sensory organs during multisensory integration and how the CNS dynamically updates these weights. Methods: In this study, we built an experimental setup that utilizes the natural refuge tracking behavior of Apteronotus albifrons, a species of weakly electric fish. These fish hide inside refuges and track their movements when needed due to their innate shelter-seeking behavior. To achieve this, Apteronotus constantly integrates visual and electrosensory signals. Our experimental setup utilizes the natural refuge tracking behavior of these fish. It consists of a refuge that moves on a single axis and a camera that captures fish movements. We used a translucent refuge to generate electrosensory signals in our experimental setup. In the dark, movements of this refuge create electrosensory signals for the fish. To generate visual cues, we used a mini projector attached to the refuge. This allows us to deliver synchronized or independent visual and electrosensory cues to the fish. Results: In our study, we experimented with N=5 fish with synchronous and independent sensory stimuli. Our results showed that electrosensory stimulation was sufficient to stimulate the refuge-tracking behavior but not the visual stimuli. Conclusion: We are currently investigating the dynamic sensory reweighting by gradually adding different frequency stimulations to model the multisensory integration process in these fish. This way, we aim to reveal the cost functions associated with sensory reweighting during multisensory integration. [ABSTRACT FROM AUTHOR]

Objective: Weakly electric fish exhibit shelter seeking behavior via hiding inside refuges in their natural habitats. These fish benefit from visual and electrosensory cues while exhibiting this behavior. Our goal is to examine the effects of the sensory salience on the refuge tracking behavior. This will help to understand the relation between sensory salience and behavioral performance. Methods: We used an experimental setup with a 3D-printed refuge, which is moved via complex stimulation trajectories using a linear actuator. We track the refuge and the fish using image processing and estimate its frequency response. We experimented with N=5 Apteronotus albifrons under different sensory conditions considering the sample sizes used in the literature. We repeated the experiments for 54 sensory conditions; illumination: dark, loess, and light, refuge structure: with/out windows, refuge length: 7, 14, 21 cm, conductivity: low, medium, and high. We used N-way ANOVA to investigate the effects of sensory salience on the tracking performance of the fish. Results: Tracking performance of the fish is evaluated based on the gain and phase lag between the stimulus and the fish. Ideally, the gain should 1, corresponding to equal amplitude, and the phase lag should be 0, corresponding to a delay-free tracking. Our results showed that illumination plays a critical role in the tracking performance as the performance degrades (lower gain, increased phase alg) in the dark. The length and structure of the refuge also have significant effects on the tracking response of the fish. Increasing the water conductivity deteriorated the tracking performance by adding additional phase lag to the response of the fish. Conclusion: In this study, we manipulated the salience of the sensory information for the two major sensory organs used by the weakly electric fish during refuge tracking. As a future work, we plan to reveal the effects of sensory salience on the multisensory integration by modeling the sensory perception of weakly electric fish. [ABSTRACT FROM AUTHOR]

Objective: The sensorimotor control processes in animals work as a closed-loop feedback control system. As a result, stimulus and behavioral responses only allow us to identify the closed-loop system dynamics, but it does not give enough information to identify neural controllers or the locomotor dynamics. We aim to estimate a proxy of the output of the neural controller of weakly electric fish during refuge tracking behavior. Our goal is to utilize deep learning techniques to track the movements of the anal fin and the nodal point. Methods: We used an experimental setup to stimulate the refuge tracking response of the weakly electric fish by moving the refuge sinusoidally at certain frequencies (0.1 Hz, 0.55 Hz, 1.15 Hz). We recorded the behavioral response of N=3 fish (back and forth movements) at 100 Hz for 30 seconds. These fish swim forward and backward by applying thrust via generating counter propagating waves along their long anal fin. The fish modulates the location, termed nodal point, where these waves meet to control the thrust it applies in either direction. In this work, we use deep learning to track the movements of the anal fin and later the movements of the nodal point. As a result, we obtain a readout of the output of the neural controller using kinematic measurements from the closed-loop control system. Results: Deep learning techniques yield successful results in tracking the anal fins. However, tracking the nodal point requires measurements from multiple (5-10) frames. The estimated trajectories of the nodal point is correlated with the velocity of the fish. Conclusion: These results show that kinematic tracking of nodal points serves as an interim signal between the controller (neural control) and plant (locomotor dynamics), which allows estimating the dynamics of the neural controller and the locomotor dynamics. In our future work, we plan to develop a realtime algorithm to track the nodal point online. [ABSTRACT FROM AUTHOR]

Objective: Animals routinely exhibit active sensing movements during their behavioral tasks to enhance the sensory feedback perceived from the environment. Understanding the behavioral mechanisms of this phenomenon necessitates separating the active sensing movements from the task-related movements. This requires hand labeling the data. In this study, we use machine learning to automatically separate active sensing movements from task-related movements. Besides, we plan to further classify the types of active sensing movements adopted by the animals. Methods: To achieve our objectives, we use a dataset about the refuge tracking response of Eigenmannia virescens. Dataset consists of 710 experimental trials collected under various sensory saliency conditions. Each trial includes fish's response to the movement of a sinusoidally moving PVC refuge in a single linear dimension. The movements of the fish, in response to refuge movements, are divided into epochs, separated by the zero crossings of the fish velocity. Then, we used the 'Bagged Trees' algorithm to train a classifier that separates the active sensing movements from the task-related movements of the fish. Results: We used 5-fold cross-validation to assess the classification performance of our model. Our results indicate that Bagged trees can detect active sensing movements with 91% accuracy on the test data. Besides, we classified the active sensing epochs into various stereotyped movements, such as scanning, drift, reset and wiggle. Our classifier can also separate the active sensing movements into these classes with 89% accuracy-Conclusion: Despite promising classification results with 91% accuracy, there is still an important gap toward automatic clustering of behavioral movements. In the future, we plan to generate state transition diagrams of the fish that quantify transition probabilities from one type of movement to the others. We will use this probabilistic information to boost the classification results by modulating the scores assigned to different classes. [ABSTRACT FROM AUTHOR]

Objective: Sensory and motor systems are linked: motor actions alter the information that animals receive from their sensory receptors, which in turn is used to control subsequent motor actions. In active sensing, animals execute ancillary motor movements to obtain and/or modulate sensory information. The neural mechanisms for the control of active sensing are not known. Our goal is to describe the neurophysiological interactions between the sensory and motor circuits in the CNS toward understanding the neural mechanisms of the active sensing. Methods: Weakly electric fishes such as Apteronotus lep-torhynchus robustly perform an image stabilization task in which animals track the movement of a refuge in a single linear dimension. During this behavior, these animals produce large fore-aft movements for active sensing that modulate and maintain a robust level of sensory slip, the main form of feedback used in the control of refuge tracking. We performed chronic tetrode recordings in midbrain circuits of (n=17) Apteronotus during their free refuge tracking behavior to reveal the sensorimotor activity related to the control of these active sensing movements. We implanted electrodes in the Torus semicircularis, a midbrain electrosensory area where direction-selective responses first emerge. Results: We found direction-selective neurons, which respond to a range of velocities for the sensory slip, the error between the refuge and fish movements. Our analysis revealed that these direction-selective neurons are also selective for ranges of acceleration of the sensory slip. Midbrain electrosensory neurons responded best to high-frequency features, behaving as a high-pass filter on the sensory slip. Conclusion: The behavioral studies suggest high-pass sensorimotor transformations in the CNS. The spatiotemporal filtering properties of the midbrain neurons we obtained matched the predictions of the behavioral experiments. These preliminary findings are a first step toward understanding neural mechanisms for control of active sensing in a freely behaving animal. [ABSTRACT FROM AUTHOR]

Objective: The behavioral performance of an individual animal is determined via an interplay between its plant and controller. The plant typically includes musculoskeletal components that interact with the environment to generate movement. The controller typically includes sensory systems and neural circuits used to process information to generate motor commands. Our goal is to describe the behavioral and neural representations of the sensorimotor transformation occurring in the controllers of the animals during their free, unconstrained behavior. Methods: Weakly electric fishes, such as Eigenmannia virescens and Apteronotus leptorhynchus, robustly perform an image stabilization task in which animals track the movement of a refuge in a single linear dimension. We studied refuge tracking in these fishes (n=5 Eigenmannia and n=17 Apteronotus) using their behavioral responses and chronic tetrode recordings to reveal the dynamics of the sensorimotor activity in the brain. Results: We identified the frequency response functions of the controllers for each fish using their behavioral responses. Despite the differences in their dynamics, each of these controllers had high-pass filtering characteristics. Moreover, analyzing the chronic tetrode recordings revealed direction-selective neurons, which respond to a range of velocities for the sensory slip, which is the error between the refuge and fish movements. The midbrain electrosensory neurons responded best to high-frequency features, behaving as a high-pass filter on the sensory slip. Conclusion: The spatiotemporal filtering properties of the midbrain neurons obtained through chronic tetrode recordings matched the system identification results of the behavioral experiments. These results suggest that sensorimotor transformation in the central nervous system pays more attention to high-frequency stimuli rather than the low-frequency content. We believe that this feature plays a critical role behind the high-frequency active sensing movements, which modulates the sensory information in the environment via moving the sensory receptors. [ABSTRACT FROM AUTHOR]