Figure 8From: Learning search polices from humans in a partially observable contextRisk-prone and risk-averse searches (red and green trajectories). Top left: Two human trajectories taken from the data shown in Figure 7. Top right: Two greedy trajectories. Bottom left: GMM trajectories, all starting from the same location; the colour coding is to illustrate the different policies which were encoded and emerge given the same initial conditions. Bottom right: Corresponding expected features of each trajectory. The colour coding matches the trajectories to the ‘GMM risk types’ sub-figure. All the searches which were generated by the GMM for this initialisation produced risk-averse searches (based on the feature metric discussed previously).Back to article page