Online learning for orientation estimation during translation in an insect ring attractor network

Abstract

Insect neural systems are a promising source of inspiration for new navigation algorithms, especially on low size, weight, and power platforms. There have been unprecedented recent neuroscience breakthroughs with Drosophila in behavioral and neural imaging experiments as well as the mapping of detailed connectivity of neural structures. General mechanisms for learning orientation in the central complex (CX) of Drosophila have been investigated previously; however, it is unclear how these underlying mechanisms extend to cases where there is translation through an environment (beyond only rotation), which is critical for navigation in robotic systems. Here, we develop a CX neural connectivity-constrained model that performs sensor fusion, as well as unsupervised learning of visual features for path integration; we demonstrate the viability of this circuit for use in robotic systems in simulated and physical environments. Furthermore, we propose a theoretical understanding of how distributed online unsupervised network weight modification can be leveraged for learning in a trajectory through an environment by minimizing orientation estimation error. Overall, our results may enable a new class of CX-derived low power robotic navigation algorithms and lead to testable predictions to inform future neuroscience experiments.

Estimating orientation in Natural scenes: A Spiking Neural Network Model of the Insect Central Complex

“…142 PEN-a neurons conjunctively encode both head direction and angular velocity [10,11]. 143 By biasing current input representing angular velocity to one half of the PB we can 144 disrupt the balance between input to EPG cells and drive the bump either clockwise of 145 anticlockwise around the ring [20][21][22][23]25]. Constant angular velocity (corresponding to 146 rotating in one direction at a constant speed) is provided to PEN cells using a GeNN 147 current source model [40] with constant magnitude.…”

“…and most other previous models are biased towards the Drosophila circuit [10,22,25,26]. However the real power of computational approaches is that they allow us to explore questions about other insect species and investigate how behavioural specialisms may come about from this conserved circuitry.…”

“…Recent Drosophila connectome data has revealed self-recurrent connections between excitatory EPG cells that likely ensure bump persistence [13, 14, 26]. This is in contrast to proposals in previous models where reciprocal connections between EPG cells and PEG cells would maintain the bump [20–23, 25, 27], however PEG to EPG connections are actually very weak [14]. Global inhibition is provided by two further cells types: Δ7 which functionally deliver strong inhibitory input to EPG cells on the ring that are opposite from the location of the bump ([13]; Fig 1D), and GABAergic ring neurons which have characteristic ring-shaped axonal projections enabling inhibition of EPG cells around the EB ([28]; Fig 1E).…”

“…The stability of a single bump relies on interactions between an excitatory population with recurrent connections and global inhibition which prevents runaway activity. Several computational models have been proposed to demonstrate how this circuit can use attractor dynamics to encode heading direction [10,[20][21][22][23][24][25][26][27]. These models all include EPG and PEN cells (or equivalent) as the excitatory component, with reciprocal connections between the two populations forming the bump by propagating activity around the ring network (Fig 1C).…”

Exploiting large neuroimaging datasets to create connectome-constrained approaches for more robust, efficient, and adaptable artificial intelligence

“…This model was demonstrated for low complexity sensor fusion of visual landmark position and angular velocity measurements on a robotic platform. 43 While not yet as performant as traditional state estimation approaches, the approach taken here has given additional insight into new classes of connectivity in this circuit and a computational platform to explore the level of biological detail required to model this circuit for sensor fusion.…”