Fig. 6: A simple vehicle model incorporating a tunable “ON” signal reproduces flexible heat avoidance behavior.

a-d An in silico Braitenberg vehicle model based on the dimensions of the fly, incorporating a simple “brain”: a flexible central processing step designed to mimic an ON response tracking stimulus rate, acting in parallel to the relay of signal from the sensor to the motor (red). a Basic model design, with key parameters used as substrate for evolution (s = sensory input, v = wheel velocity; parameters: \(a\) = gain, \(b\) = offset, \(\varepsilon,\gamma\) = noise, \({w}_{j},\, {w}_{c}\) = weights of ipsi- and contralateral connections, respectively, \(m\) = multiplier, \({\tau }_{p}\) = time constant). b Brain design. c Conceptualization of the transformation performed by the filter for signals of different rate (\(s\) = input signal, \(g\) = output) (d) Schematic of the evolutionary process used to optimize the parameters. Only the multiplier, \(m\), is allowed to vary between experimental conditions (25° vs 30 °C, 25° vs 35 °C and 25° vs 40 °C). e Input-output transformation at the central processing step following evolution, applied to representative signals of different rate (s = input, g = output). f Error convergence for the five different objectives as a result of evolution (median ± median absolute deviation; error values are from each generation’s Pareto front vehicles; the error of each vehicle is normalized by the median error of the final Pareto front vehicles). g 3D scatter plot showing the error space for 3 key objectives for all vehicles tested (gray), the all-time best performing vehicles after 500 generations (red), and a chosen top performer (green, chosen at random from a group of top performers, see methods for details). X-axis = Crossover/U-turn ratio error, Y-axis = early turn frequency error, Z-axis = Left/Right turn predictability error. h–j The evolved top performing vehicle (green dot in g) nearly reproduces fly thermotactic behavior in a simulated arena. h Vehicle trajectories in the simulated arena (25° vs 40 °C, arrowhead = start). i,j Vehicle performance parameters (n = 400 simulations), i: boxplots: black line = median, box = interquartile range, whiskers = range. k–n A flexible “brain” is essential for the vehicle to reproduce the appearance of early turns in the 25° vs 35 °C and 25° vs 40 °C conditions. k–m Quantification of turning frequency within distinct bands of the thermal gradients in (k) real flies (N = [25,24,25] for 25° vs 30°,35°,40 °C, respectively), (l) a top performing evolved Braitenberg vehicle and (m) a top performing evolved vehicle in which the parameter, m, is allowed to reach 3 different solutions for the 3 experimental conditions. Only the latter vehicle displays appropriate early turns, similar to the real fly, and (n) can minimize early turn error; boxplots: orange line = median, box = interquartile range, whiskers = range (n = 1056 top performers). o Across evolutionary time the influence of the central filter (parameter m) settles on values proportional to the frequency of early turns in each experimental condition (n = 112 vehicles/generation, median ± median absolute deviation). p, q Independently increasing the value of m (i.e., the influence of the central processing step) increases the frequency of early turns (n = 200 simulations each). Source data are provided as a source data file.