guest ::
| Home > Publications database > Finding the Goal: Insect-Inspired Spiking Neural Network for Heading Error Estimation |
| Home > Publications database > Finding the Goal: Insect-Inspired Spiking Neural Network for Heading Error Estimation |
| Contribution to a conference proceedings | FZJ-2024-01293 |
;
2023
IEEE
This record in other databases: 
Please use a persistent id in citations: doi:10.1109/IROS55552.2023.10342210
Abstract: Insects have extraordinary navigational abilities.Monarch butterflies migrate every year to the same forestover hundreds of kilometers, desert ants find their way backto the nest tens of meters away and dung beetles maintainthe same heading direction over meters. The performance ofthese agents has been optimized by evolution over the last500 million years leading to power-efficient, low-latency andprecise sensorimotor systems. Research efforts in the fieldof neuroscience, biology and robotics are instrumental foruncovering the neural substrate of insect navigation abilities.The development of models of insect navigation tightly coupledwith the insect connectome and neurophysiology and theirembedding in closed loop systems support the understanding ofembodied animal cognition and can advance robotic systems.In this work, we focus on insect navigation because of theefficient insect navigational apparatus. Furthermore, the recentdiscovery of the central complex, the neuronal center of insectnavigation, facilitates the development of new hypotheses aboutinsect navigation. All navigating insects need to perform somekind of goal-directed behavior during which they have toreach a specific goal location or maintain the same movementdirection over long distances. Such behavior requires theagent to be aware of its current heading direction, desiredheading direction, and the error between them. Building onprevious research in the field, we propose a novel model forthis error estimation that can in principle be generalized forall navigating insect species. We implement the model in aspiking neural network and test its capabilities on a simulatedrobotic platform. The precision of the network is comparableto or even better than the biological role model. Thus, ourimplementation serves as a working hypothesis for how theheading error might be computed in the insect brain. Our modelwill help to explain navigational behavior in fruit flies, orchidbees, bumble bees and some less researched insect species.Furthermore, its simplicity in comparison to other models andimplementation in a spiking neural network makes it verysuitable for neuromorphic systems, an emerging field of braininspired hardware.
|
The record appears in these collections: |
PDF