Fig. 2: Overview of the route-centric lateralized memories model implemented in the Antcar robot.
From: Route-centric ant-inspired memories enable panoramic route-following in a car-like robot

This figure illustrates the processing pipeline from image encoding to navigation control, spanning both the learning and exploitation phases. a The Antcar robot: a compact car-like platform equipped with a panoramic camera and a GPS-RTK (Global Positioning System - Real-Time Kinematic) system for ground truth data. b Image encoding: this process mimics the ant’s visual system. Panoramic images (I) are captured, blurred, sub-sampled, and edge-filtered to produce a low-resolution 32 × 32 pixel panorama (IS). The resulting image is then transformed into Projection Neurons (PN), expanded into Excitatory Post-Synaptic Projections (EP), and reduced into Action Potentials (AP) via a κ-WTA function, forming the sparse Kenyon Cell (KC) representation. c Learning phase: the robot follows a route (C) from a start point (N). An in silico scan (simulated image rotation) generates images with a defined angular error (\(\hat{{\theta }_{{{\rm{e}}}}}\)), used to continuously categorize views at 38 Hz in a route-centric manner---i.e., as facing to the right or left of the local route orientation. These continuous categorizations drive the update of synaptic weights in the corresponding Mushroom Body Output Neurons (MBONs), allowing visual inputs to be associated with specific route memories in a self-supervised fashion. Joystick inputs are used to define learning boundaries at the start and end of the route, modulating plasticity. d Exploitation phase: the robot seeks to minimize lateral (d) and angular (θe) errors with respect to the learned route. The encoded image continuously activates MBONs at 16Hz based on previously learned synaptic weights, enabling the robot to infer route orientation and adjust its steering and speed accordingly. MBON familiarity indices (λ) operate in an opponent process: differences in familiarity between left and right MBONs guide steering, while the overall familiarity magnitude modulates speed. Special MBONs associated with route extremities affect motivational states, enabling route polarity correction or halting movement.