Fig. 6: Sweeps cover nearby space with optimal efficiency.
From: Left–right-alternating theta sweeps in entorhinal–hippocampal maps of space
a, Illustration of an autonomous sweep-generating artificial agent. The agent is shown at a series of three positions as it moves from left to right. At each time step, it generates a beam-shaped ‘sweep’ (white illuminated regions) in a particular direction (blue arrows). The sweep footprints persist in time and serve as a memory trace of previously covered locations. At the current time step (right), the agent calculates the direction that minimizes overlap with the coverage trace; this will become the direction of its next sweep. b, The agent is moved along a scale-free linear path (grey horizontal line) and generates a sweep at each time step (sweeps as shown in a). The selected sweep directions (blue arrows) spontaneously alternate between two directions relative to the direction of movement. c, Alternation of sweep directions in 1,000 runs of the simulation in b. Circles show the mean alternation score (range 0–1) for each time step; error bars indicate the 5th and 95th percentiles across runs. The horizontal line indicates the expected alternation score for random, uniformly distributed angles. d, Illustration of an alternative ‘empirically driven’ agent that predicts observed sweep directions. As in a, the agent chooses the overlap-minimizing sweep direction at every time step. However, here the sweeps in the previous coverage trace are set at empirically decoded positions and directions, instead of the agent’s previous choices. Hence, this model variant predicts the optimal next sweep direction, given the empirical data up to the current time. e, Sweep directions predicted by the empirically driven agent shown in d (blue shapes) and internal direction fitted by the LMT model (green arrows) during 16 theta cycles in an open field (both plotted in a head-centred reference frame). Theta cycles are evenly spaced along the horizontal axis. Note the alignment between decoded and predicted directions. f, Running speed modulates the distribution of sweep direction. Top row, distribution of head-centred sweep directions chosen at different running speeds by the agent in an open-field session (same session as in e, but the simulation was now run in ‘self-driving’ mode without the influence of decoded direction). Bottom row, distributions of head-centred internal direction decoded from empirical data (in the same session). In both cases, the bimodality of the distribution increases with running speed. g, Alternation of sweep directions increases with speed. Left, average alternation score of the sweep directions chosen at different speeds by the agent in self-driving mode during open-field foraging. Each set of connected coloured dots shows data from a different rat (n = 13). The horizontal line indicates the expected alternation score for random, uniformly distributed angles. Right, the same as the left graph, except using decoded (empirical) internal direction (with the same animals). Credits: rats, scidraw.io/Gil Costa; robots, openclipart.org/annares.
