Abstract
This paper discusses trust in situation-adaptive autonomy (SAA), where the automation may change its level dynamically and it can have the authority of decision making under time critical situation. Although SAA is necessary and effective for attaining systems safety, it is not obvious whether or not humans accept SAA. We analyze subjective rating of trust which subjects have in the efficacy of the automation, and show the results of four kinds of analyses on trust. First, an ANOVA indicates that trust is high (low) when the reliability of the automation is high (low). Second, a first order lag model of trust clarifies the following two points: (1) Trust decreases when the automation behaves erroneously, (2) Trust declines when a human disagrees with the automation even when the automation is correct. Third, an analysis of relationship between changes of trust and the reliability of automation shows the following results: (a) Trust has a tendency to stay the same if the automation is completely reliable, (b) Subjects tend to change the level of trust if the automation is reliable but imperfect. Finally, we analyze trust under the situation where the automation performs safety control actions without allowing human intervention. The results have been obtained: (i) Trust does not decline if the reliability of the automation is 100%, (ii) Subjects tend to decrease their ratings of trust when the automation is highly but not entirely reliable. The above results suggest that situation awareness of a human operator is an important factor to make SAA acceptable to the operator.
