With the advent of highly automated cars (SAE level 3), new uses are emerging. The driver can delegate the control of the vehicle to an automation system and may engage in non-driving activities. As a consequence, the driver may withdraw not only from the physical control of the vehicle but also from the monitoring of the driving environment. In this study, we examined the impact of non-driving activity on gaze data, cardiac data and the quality of takeover. We collected 66 trials in an autonomous driving simulator (SAE level 3) in two mental workload conditions induced by two distinct task-difficulty levels. The results showed an effect of the non-driving task difficulty on mental workload measured with the NASA-TLX but not on cardiac data. Visual strategies were also influenced by the difficulty levels but not by the quality of the takeover. However, the orientation of gaze at the time of takeover request appeared to determine the quality of takeover , with better performance observed for driver looking at the road scene compared to distracted drivers.