Whisker Movements Reveal Spatial Attention: A Unified Computational Model of Active Sensing Control in the Rat
Figure 3
Boxes indicate components, solid arrows indicate causal influences. Extends/modifies model of Figure 2 with implementation-specific components: attended region made explicit as salience map(s); ‘other signals’ implemented as an endogenous stochastic source; inhibition, including a contribution from inhibition-of-return (IOR) system; oscillator (OSC); ‘physical plane model’ simulates mechanics. Separate salience channels are maintained for tactile and ‘other’ signals and selected at switch (SW). Data snapshot. Within the boxes is displayed a data snapshot from a point of maximum protraction during a whisk against a vertical surface. Physical plane model in world-centric coordinates includes head, whiskers and obstacle surfaces; whisker contacts are shown as dots (darker dots indicate stronger bending) and current target of foveation as a ‘target’ icon. Tactile and ‘other’ signals are mapped into head-centric excitation maps which drive salience maps (darker areas represent higher salience; pattern corresponding to wall can be seen in tactile salience map). Activity in salience map regions inside obstacles as well as in previously-visited regions (IOR, see text) is inhibited. Tactile salience channel is selected at SW owing to higher peak salience than ‘other’ channel. Whisker movement panel shows maximum protraction computed to roughly achieve MIMC with respect to attended region. Head movement panel shows current target of foveation (target icon) at peak of salience map. Video S4 shows the operation of the implementation during a trial including this snapshot (which was taken at t = 0.340).