Looming detection by identified visual interneurons during larval development of the locust Locusta migratoria

A Robust Collision Perception Visual Neural Network With Specific Selectivity to Darker Objects

Building an efficient and reliable collision perception visual system is a challenging problem for future robots and autonomous vehicles. The biological visual neural networks, which have evolved over millions of years in nature and are working perfectly in the real world, could be ideal models for designing artificial vision systems. In the locust's visual pathways, a lobula giant movement detector (LGMD), that is, the LGMD2, has been identified as a looming perception neuron that responds most strongly to darker approaching objects relative to their backgrounds; similar situations which many ground vehicles and robots are often faced with. However, little has been done on modeling the LGMD2 and investigating its potential in robotics and vehicles. In this article, we build an LGMD2 visual neural network which possesses the similar collision selectivity of an LGMD2 neuron in locust via the modeling of biased-ON and -OFF pathways splitting visual signals into parallel ON/OFF channels. With stronger inhibition (bias) in the ON pathway, this model responds selectively to darker looming objects. The proposed model has been tested systematically with a range of stimuli including real-world scenarios. It has also been implemented in a micro-mobile robot and tested with real-time experiments. The experimental results have verified the effectiveness and robustness of the proposed model for detecting darker looming objects against various dynamic and cluttered backgrounds.

Retinal ganglion cell defects cause decision shifts in visually evoked defense responses

A variety of visual cues can trigger defensive reactions in mice and other species. In mice, looming stimuli that mimic an approaching aerial predator elicit flight or freezing reactions, while sweeping stimuli that mimic an aerial predator flying parallel to the ground typically elicit freezing. The retinal ganglion cell (RGC) types involved in these circuits are largely unknown. We previously discovered that loss of RGC subpopulations in Brn3b knockout mice results in distinct visual response deficits. Here, we report that retinal or global loss of Brn3b selectively ablates the fleeing response to looming stimuli while leaving the freeze response intact. In contrast, freezing responses to sweeping stimuli are significantly affected. Genetic manipulations removing three RGC subpopulations (Brn3a⁺ betta RGCs, Opn4⁺Brn3b⁺, and Brn3c⁺Brn3b⁺ RGCs) result in milder phenocopies of Brn3b knockout response deficits. These findings show that flight and freezing responses to distinct visual cues are mediated by circuits that can already be separated at the level of the retina, potentially by enlisting dedicated RGC types.NEW & NOTEWORTHY Flight and freezing response choices evoked by visual stimuli are controlled by brain stem and thalamic circuits. Genetically modified mice with loss of specific retinal ganglion cell (RGC) subpopulations have altered flight versus freezing choices in response to some but not other visual stimuli. This finding suggests that "threatening" visual stimuli may be computed already at the level of the retina and communicated via dedicated pathways (RGCs) to the brain.

Molecular characterization and distribution of the voltage-gated sodium channel, Para, in the brain of the grasshopper and vinegar fly

Voltage-gated sodium (NaV) channels, encoded by the gene para, play a critical role in the rapid processing and propagation of visual information related to collision avoidance behaviors. We investigated their localization by immunostaining the optic lobes and central brain of the grasshopper Schistocerca americana and the vinegar fly Drosophila melanogaster with an antibody that recognizes the channel peptide domain responsible for fast inactivation gating. NaV channels were detected at high density at all stages of development. In the optic lobe, they revealed stereotypically repeating fascicles consistent with the regular structure of the eye. In the central brain, major axonal tracts were strongly labeled, particularly in the grasshopper olfactory system. We used the NaV channel sequence of Drosophila to identify an ortholog in the transcriptome of Schistocerca. The grasshopper, vinegar fly, and human NaV channels exhibit a high degree of conservation at gating and ion selectivity domains. Comparison with three species evolutionarily close to Schistocerca identified splice variants of Para and their relation to those of Drosophila. The anatomical distribution of NaV channels molecularly analogous to those of humans in grasshoppers and vinegar flies provides a substrate for rapid signal propagation and visual processing in the context of visually-guided collision avoidance.

Towards Computational Models and Applications of Insect Visual Systems for Motion Perception: A Review

Motion perception is a critical capability determining a variety of aspects of insects' life, including avoiding predators, foraging, and so forth. A good number of motion detectors have been identified in the insects' visual pathways. Computational modeling of these motion detectors has not only been providing effective solutions to artificial intelligence, but also benefiting the understanding of complicated biological visual systems. These biological mechanisms through millions of years of evolutionary development will have formed solid modules for constructing dynamic vision systems for future intelligent machines. This article reviews the computational motion perception models originating from biological research on insects' visual systems in the literature. These motion perception models or neural networks consist of the looming-sensitive neuronal models of lobula giant movement detectors (LGMDs) in locusts, the translation-sensitive neural systems of direction-selective neurons (DSNs) in fruit flies, bees, and locusts, and the small-target motion detectors (STMDs) in dragonflies and hoverflies. We also review the applications of these models to robots and vehicles. Through these modeling studies, we summarize the methodologies that generate different direction and size selectivity in motion perception. Finally, we discuss multiple systems integration and hardware realization of these bio-inspired motion perception models.

Ontogeny and development of the tritocerebral commissure giant (TCG): an identified neuron in the brain of the grasshopper Schistocerca gregaria

The tritocerebral commissure giant (TCG) of the grasshopper Schistocerca gregaria is one of the best anatomically and physiologically described arthropod brain neurons. A member of the so-called Ventral Giant cluster of cells, it integrates sensory information from visual, antennal and hair receptors, and synapses with thoracic motor neurons in order to initiate and regulate flight behavior. Its ontogeny, however, remains unclear. In this study, we use bromodeoxyuridine incorporation and cyclin labeling to reveal proliferative neuroblasts in the region of the embryonic brain where the ventral giant cluster is located. Engrailed labeling confirms the deutocerebral identity of this cluster. Comparison of soma locations and initial neurite projections into tracts of the striate deutocerebrum help identify the cells of the ventral cluster in both the embryonic and adult brain. Reconstructions of embryonic cell lineages suggest deutocerebral NB1 as being the putative neuroblast of origin. Intracellular dye injection coupled with immunolabeling against neuron-specific horseradish peroxidase is used to identify the VG1 (TCG) and VG3 neurons from the ventral cluster in embryonic brain slices. Dye injection and backfilling are used to document axogenesis and the progressive expansion of the dendritic arbor of the TCG from mid-embryogenesis up to hatching. Comparative maps of embryonic neuroblasts from several orthopteroid insects suggest equivalent deutocerebral neuroblasts from which the homologous TCG neurons already identified in the adult brain could originate. Our data offer the prospect of identifying further lineage-related neurons from the cluster and so understand a brain connectome from both a developmental and evolutionary perspective.

Two identified looming detectors in the locust: ubiquitous lateral connections among their inputs contribute to selective responses to looming objects

In locusts, two lobula giant movement detector neurons (LGMDs) act as looming object detectors. Their reproducible responses to looming and their ethological significance makes them models for single neuron computation. But there is no comprehensive picture of the neurons that connect directly to each LGMD. We used high-through-put serial block-face scanning-electron-microscopy to reconstruct the network of input-synapses onto the LGMDs over spatial scales ranging from single synapses and small circuits, up to dendritic branches and total excitatory input. Reconstructions reveal that many trans-medullary-afferents (TmAs) connect the eye with each LGMD, one TmA per facet per LGMD. But when a TmA synapses with an LGMD it also connects laterally with another TmA. These inter-TmA synapses are always reciprocal. Total excitatory input to the LGMD 1 and 2 comes from 131,000 and 186,000 synapses reaching densities of 3.1 and 2.6 synapses per μm² respectively. We explored the computational consequences of reciprocal synapses between each TmA and 6 others from neighbouring columns. Since any lateral interactions between LGMD inputs have always been inhibitory we may assume these reciprocal lateral connections are most likely inhibitory. Such reciprocal inhibitory synapses increased the LGMD’s selectivity for looming over passing objects, particularly at the beginning of object approach.

Background visual motion affects responses of an insect motion-sensitive neuron to objects deviating from a collision course

Stimulus complexity affects the response of looming sensitive neurons in a variety of animal taxa. The Lobula Giant Movement Detector/Descending Contralateral Movement Detector (LGMD/DCMD) pathway is well-characterized in the locust visual system. It responds to simple objects approaching on a direct collision course (i.e., looming) as well as complex motion defined by changes in stimulus velocity, trajectory, and transitions, all of which are affected by the presence or absence of background visual motion. In this study, we focused on DCMD responses to objects transitioning away from a collision course, which emulates a successful locust avoidance behavior. We presented each of 20 locusts with a sequence of complex three-dimensional visual stimuli in simple, scattered, and progressive flow field backgrounds while simultaneously recording DCMD activity extracellularly. DCMD responses to looming stimuli were generally characteristic irrespective of stimulus background. However, changing background complexity affected, peak firing rates, peak time, and caused changes in peak rise and fall phases. The DCMD response to complex object motion also varied with the azimuthal approach angle and the dynamics of object edge expansion. These data fit with an existing correlational model that relates expansion properties to firing rate modulation during trajectory changes.

Excitation and inhibition in recurrent networks mediate collision avoidance in Xenopus tadpoles

Information processing in the vertebrate brain is thought to be mediated through distributed neural networks, but it is still unclear how sensory stimuli are encoded and detected by these networks, and what role synaptic inhibition plays in this process. Here we used a collision avoidance behavior in Xenopus tadpoles as a model for stimulus discrimination and recognition. We showed that the visual system of the tadpole is selective for behaviorally relevant looming stimuli, and that the detection of these stimuli first occurs in the optic tectum. By comparing visually guided behavior, optic nerve recordings, excitatory and inhibitory synaptic currents, and the spike output of tectal neurons, we showed that collision detection in the tadpole relies on the emergent properties of distributed recurrent networks within the tectum. We found that synaptic inhibition was temporally correlated with excitation, and did not actively sculpt stimulus selectivity, but rather it regulated the amount of integration between direct inputs from the retina and recurrent inputs from the tectum. Both pharmacological suppression and enhancement of synaptic inhibition disrupted emergent selectivity for looming stimuli. Taken together these findings suggested that, by regulating the amount of network activity, inhibition plays a critical role in maintaining selective sensitivity to behaviorally-relevant visual stimuli.

Individual pause-and-go motion is instrumental to the formation and maintenance of swarms of marching locust nymphs

The principal interactions leading to the emergence of order in swarms of marching locust nymphs was studied both experimentally, using small groups of marching locusts in the lab, and using computer simulations. We utilized a custom tracking algorithm to reveal fundamental animal-animal interactions leading to collective motion. Uncovering this behavior introduced a new agent-based modeling approach in which pause-and-go motion is pivotal. The behavioral and modeling findings are largely based on motion-related visual sensory inputs obtained by the individual locust. Results suggest a generic principle, in which intermittent animal motion can be considered as a sequence of individual decisions as animals repeatedly reassess their situation and decide whether or not to swarm. This interpretation implies, among other things, some generic characteristics regarding the build-up and emergence of collective order in swarms: in particular, that order and disorder are generic meta-stable states of the system, suggesting that the emergence of order is kinetic and does not necessarily require external environmental changes. This work calls for further experimental as well as theoretical investigation of the neural mechanisms underlying locust coordinative behavior.

A look into the cockpit of the developing locust: looming detectors and predator avoidance

For many animals, the visual detection of looming stimuli is crucial at any stage of their lives. For example, human babies of only 6 days old display evasive responses to looming stimuli (Bower et al. [1971]: Percept Psychophys 9: 193-196). This means the neuronal pathways involved in looming detection should mature early in life. Locusts have been used extensively to examine the neural circuits and mechanisms involved in sensing looming stimuli and triggering visually evoked evasive actions, making them ideal subjects in which to investigate the development of looming sensitivity. Two lobula giant movement detectors (LGMD) neurons have been identified in the lobula region of the locust visual system: the LGMD1 neuron responds selectively to looming stimuli and provides information that contributes to evasive responses such as jumping and emergency glides. The LGMD2 responds to looming stimuli and shares many response properties with the LGMD1. Both neurons have only been described in the adult. In this study, we describe a practical method combining classical staining techniques and 3D neuronal reconstructions that can be used, even in small insects, to reveal detailed anatomy of individual neurons. We have used it to analyze the anatomy of the fan-shaped dendritic tree of the LGMD1 and the LGMD2 neurons in all stages of the post-embryonic development of Locusta migratoria. We also analyze changes seen during the ontogeny of escape behaviors triggered by looming stimuli, specially the hiding response.