Supporting take-over situations in highly automated driving

Highly automated vehicles (HAV) will have an enormous impact on drivers’ safety, efficiency, and comfort. The next generation of HAVs (i. e., Limited Self Driving Automation; SAE Level 3) will allow users to be engaged in non-driving related tasks (NDRT), absolving them of the need to visually monitor the roadway constantly. Instead, such vehicles are expected to present take-over requests in a timely manner, insofar as it would be sufficient to allow users to resume vehicle control comfortably, in order to deal with any (unexpected) scenarios that vehicle automation has not been designed to handle. Such situations, termed take-over situations, require the vehicle to present a take-over request (TOR) that notifies users to resume vehicle control. Take-over situations require users to disengage from the NDRTs, shift their attention to the driving scene, perceive and understand the context, make decisions, and perform appropriate maneuvers accordingly. Therefore, there is a critical need to support users to ensure smooth transitions from their NDRT to resuming vehicle control, whenever necessary. Thus, the design of TORs should consider the levels of situational awareness of users at take-over, as well as presentation properties of TORs that ensure fast and safe transitions. Despite that advances in computing have allowed users to perform multiple concurrent activities or switch between tasks, when interacting with machines, human cognitive capabilities have not increased, leaving us vulnerable to errors. Therefore we investigate the following research question: „How can we support a driver’s ability to seamlessly switch from engaging with a nondriving related task to monitor and resume the diverse complex maneuvers that constitute effective vehicle handling?“ To answer this question, we studied TORs from three aspects: (a) presentation information and modality of TORs, (b) situational factors that affect TOR responses, and (c) decision priming and NDRT engagement. We explored interaction concepts and prototypes of TORs in several experiments conducted in low, medium and high fidelity driving simulators. We found out that the presentation modality of TORs should adapt to the modality of NDRT to avoid conflicts sharing of mental resources. Conveying contextual information through TORs result in faster and safer responses. Users’ responses, however, are not only guided by the presentation parameters of the TORs, but also the road context and presence of motion and visual cues where TORs are issued. Furthermore, priming the users with decisions about upcoming maneuvers results in faster and safer takeover behavior independent of their level of engagement in NDRTs. Drawing on the results of four experiments, we provide implications and guidelines for the design of TORs which can be applied in future research and support industries in developing assistant systems for users of highly automated vehicles.