Effects of the level of autonomy on the human operator. As the level of autonomy increases, A) fewer instructions are required to produce the desired robot behavior, B) cognitive availability improves, and C) participants make better decisions, as found by comparing to the optimal agent’s decisions. Regret is the expected reward of the optimal agent’s decision minus the expected reward of the player’s decision (41) and is normalized here by the maximum impact of the decision on expected reward. We separate low-impact (L) and high-impact (H) decisions because the frequency of each type of decision differs for the different experimental conditions; the Materials and methods section describes how we account for this statistically. During high-impact decisions, one of the navigation options available to the participant will likely result in between 1-point and 2-point decrease in reward based on known information. Error bars indicate SE. Asterisks indicate statistical significance: ^*P ≤ 0.05, ^**P ≤ 0.01, ^***P ≤ 0.001.