Elements of a foraging task. A. During the work period, marmosets made a sequence of saccades to visual targets. A trial consisted of 3 consecutive saccades, at the end of which the subject was rewarded by a small increment of food. We tracked the eyes, the tongue, and the food. B. An example of two consecutive work-harvest periods, showing reward-relevant saccades (eye velocity) and tongue endpoint displacement with respect to the mouth. C. Data for two sessions, one where the tube was placed close to the mouth (orange trace), and one where it was placed farther away (red trace). Two types of licks are shown: inner-tube licks, and outer-tube licks. Depending on food location, both types of licks can contact the food. Data on the right two panels show endpoint displacement and velocity of the tongue during inner-tube licks. Error bars are SEM. D. During the work period, the subjects attempted about 8 trials on average, succeeding in 4-5 trials before starting harvest, and then licked about 18 times to extract the food.

Theoretical results of an optimal control policy. A. Capture rate (Eq. 1) is plotted during the harvest period as a function of lick number following various number of work trials. When the effort cost of licking is low (left plot,), the optimal work period is (red trace). When the effort cost is higher (right plot,), it is best to work longer before initiating harvest. B. The metabolic cost of licking (Eq. 2) is minimized when a lick has a specific duration. Tube distance varied from 0.1 to 0.3. Optimal duration that minimizes lick cost grows linearly with tube distance. C. Optimal number of work trials and licks as a function food tube distance. As the effort cost of harvest increases, one should respond by working longer, delaying harvest. D. Optimal lick duration as a function of food tube distance. The lick duration that maximizes the capture rate is smaller than the one that minimizes the lick metabolic cost (Fig. 2B). That is, it is worthwhile moving vigorously to acquire reward. However, grows faster than linearly as a function of tube distance. Thus, as the tube moves farther, it is best to reduce lick vigor. Hunger, modeled as increased value of reward, should promote work and increase vigor, while effort cost of harvest (tube distance) should promote work but reduce vigor. Parameter values for all simulations:,,,, (low food value, less hunger), (high food value, hungry).

As the effort cost of harvest increased, subjects chose to work more trials, but slowed their movements. A. Left: the number of trials attempted and succeeded per work period as a function of tube distance. Middle: successful trials per work period as a function of time during the recording session. Tube distance is with respect to a marker on the nose. Right: food available in the tube at the start of the harvest. B. Peak saccade velocity as a function of amplitude for reward-relevant and other saccades. C. Vigor of reward relevant saccades as a function of trial number during the work period. Saccade vigor was greater when the tube was closer. Pupil size is quantified during the same work periods. Accuracy is quantified as the magnitude of the saccade’s endpoint error vector (with respect to the target) and the variance of that error vector (determinant of the variance-covariance matrix), plotted as a function of the vigor of the saccade (bin size=0.05 vigor units). Error bars are SEM.

As the effort cost of harvest increased, lick vigor declined, and the pupils constricted. A. Peak speed of reward-seeking and grooming licks during protraction and retraction as a function of lick amplitude. B. Vigor of reward seeking licks (protraction) and pupil size as a function of lick number during harvest at various tube distances. C. Lick vigor and pupil size as a function of time during the entire recording session. Line colors depict tube distance as in part B. D. Average lick vigor and pupil size during a harvest as a function of number of trials successfully completed in the previous work period. Lick vigor and pupil size were greater when more food had been stored. E. Following a successful lick (contact with food), the next lick was more vigorous, and pupils dilated. Following a failed lick, the next lick was slowed, and pupils were less dilated. F. We observed no consistent effect of lick vigor on lick accuracy across subjects or across tube distances. Error bars are SEM.

Relatively low body weight, potentially reflecting a greater valuation of reward, coincided with longer work periods and greater vigor. Pupil size correlated with both vigor and decisions. A. Trials successfully completed during a work period as a function of normalized body weight at the start of the session. B. Left: saccade vigor as a function of trial number for low and high body weights. Right: pupil size during the same saccades. C. Left: lick vigor as a function of lick number during the harvest period. Right: pupil size during the same licks. D. Saccade vigor during the work period, and lick vigor during the harvest period, as a function of pupil size. E. Work duration and harvest duration as a function of pupil size. Error bars are SEM.

Number of licks per harvest as a function of tube distance.