Emergence of robust looming selectivity via coordinated inhibitory neural computations

TL;DR

This work addresses how robust looming selectivity emerges in visual systems by integrating four inhibitory mechanisms—feed-forward (FFI), global (GI), self (SI), and lateral (LI) inhibition—within a multi-layer, multi-scale network modeling the locust LGMD pathway. The authors introduce a novel numerical formulation where SI and LI are coordinated through FFI as an angular-size indicator, while GI normalizes excitations upstream, together producing selective responses to approaching objects and suppressing translation and other movements. A key contribution is the first numerical modelling of self-inhibition and its interaction with lateral inhibition, demonstrating that self-inhibition rapidly suppresses local excitation for small angular sizes and becomes less effective as objects enlarge, whereas LI becomes stronger with larger angular size to distinguish approach from recession. Across synthetic, real-world, coherence, and ablation experiments, the model achieves robust approaching selectivity, supports early collision alerts, and shows clear functional roles for each inhibition type, suggesting broad applicability to bio-inspired collision detection in dynamic scenes.

Abstract

In the locust's lobula giant movement detector neural pathways, four categories of inhibition, i.e., global inhibition, self-inhibition, lateral inhibition, and feed-forward inhibition, have been functionally explored in the context of looming perception. However, their combined influence on shaping selectivity to looming motion remains unclear. Driven by recent physiological advancements, this paper offers new insights into the roles of these inhibitory mechanisms at multiple levels and scales in simulations, refining the specific selectivity for responding only to objects approaching the eyes while remaining unresponsive to other forms of movement. Within a feed-forward, multi-layer neural network framework, global inhibition, lateral inhibition, self-inhibition, and feed-forward inhibition are integrated. Global inhibition acts as an immediate feedback mechanism, normalising light intensities delivered by ommatidia, particularly addressing low-contrast looming. Self-inhibition, modelled numerically for the first time, suppresses translational motion. Lateral inhibition is formed by delayed local excitation spreading across a larger area. Notably, self-inhibition and lateral inhibition are sequential in time and are combined through feed-forward inhibition, which indicates the angular size subtended by moving objects. Together, these inhibitory processes attenuate motion-induced excitation at multiple levels and scales. This research suggests that self-inhibition may act earlier than lateral inhibition to rapidly reduce excitation in situ, thereby suppressing translational motion, and global inhibition can modulate excitation on a finer scale, enhancing selectivity in higher contrast range.

Figures (17)

• Figure 1: The schematic morphology of dendrites, without metrics, includes three presynaptic fields synapsing to the LGMD. Field A receives visual movement-induced excitations, while fields B and C receive inhibitory signals. The spike initiation zone (SIZ) indicates where the LGMD generates and delivers spikes along the axon.
• Figure 2: The schematic diagram illustrates multi-level, multi-scale, concomitant inhibitions across four neuropils: retina, lamina, medulla, and lobula, interacting with excitation. Feed-forward inhibition operates globally, indicating angular size and mediating the effects of self-inhibition and lateral inhibition in the medulla. global inhibition normalizes excitations to a specific range. Self-inhibition curtails local excitation before it spreads to neighbouring fields. Lateral inhibition competes with excitation on a larger scale than self-inhibition. The remaining excitatory signals are eventually integrated by the LGMD cell.
• Figure 3: This schematic illustrates the interactions between different types of inhibition and excitation, shaping the response of the LGMD cell in visual signal processing. FFI acts globally to encode angular size and governs the time-dependent trade-off between LI and SI. GI normalizes lateral excitation upstream of LI and SI. SI initially reduces local excitation at the smallest scale. LI subsequently reduces remaining excitation. LGMD cell sieves and absorbs grouped excitations. Dynamic threshold mechanism is applied during spike initiation.
• Figure 4: The time-varying coefficient computed by FFI in Eq. \ref{['Eq:AS']} is associated with LI and SI in a complementary fashion. The red and white shading indicates the respective weights of SI and LI over time during (a) approaching stimuli, (b) translating stimuli, (c) receding stimuli, and (d) grating stimuli. The object size is defined as the ratio of the moving object’s area relative to the entire visual field.
• Figure 5: The illustrations depict coherent–incoherent dark looming stimuli within an identical time window. Here, dark looming refers to a black square expanding against a white background, representing an approaching object. The coherence degree varies from $5\%$ to $100\%$, indicating the proportion of black pixels in each frame that constitutes the looming square.