20 Décembre – Thesis defense - Jonathan Deniel

10 h Amphi - ENSC / Bordeaux INP (Talence)

Risk awareness and vehicle automation:from the analysis of cognitive processes in a lane change situation to the human-centered design of a human-like automation system.

Driving activity is both widespread and very complex. Among the most frequently executed manœuvres we identified the lane change (LC) with insertion between traffic vehicles. Current improvements in the field of driving automation are progressively paving the way for the gradual automation of the various components of the driving activity, including, among other things, lane changes.
Our purposes in this thesis work in cognitive engineering were (1) to better understanding and contribution to the modeling of the driver's cognitive processes at work when making decisions and performing LC in manual driving (especially regarding the subjective risk assessment), in order to feed the development of the gls{COSMODRIVE} model (cites{bellet2003}). (2) Design and evaluate a textit{Human-like} automation capable of executing a LC as a human driver. This second objective also makes it possible to address questions about the acceptance of the behaviour of this type of automations. (3) Study the possible effects induced by the use of this LC automation, on decisions and judgments regarding lane change situations in subsequent manual driving situations.
To achieve these objectives, we set up a research protocol (on the V-HCD driving simulation platform) divided into three intertwined experimental phases and involving the same participants for each of them. Participants were first invited to make decisions on LC and to assess the situational risk and acceptability of a hypothetical automated LC system. Then, in the second phase, they were invited to use and evaluate textit{Human-like} LC automation we designed for this experiment. Finally, in the last phase, participants were instructed to drive again manually and to make LC decisions in driving situations that were specially configured to approach their decision threshold (estimated from a direct analysis of their decisions during the first manual driving phase).
The results obtained led us to identify a traffic merging strategy within the LC manœuvre allowing us, on the basis of the envelope zones theory, to explain the structure of the LC decisions. Concerning the human-like LC automation we found a better acceptance of the system by the participants than expected, even from the most reluctants to the autonomous vehicle. We also showed a decrease in the risk estimation of the situation and the manoeuvre when it was performed by the automation. We additionnaly validated the textit{Human-like} aspect of our automation's LC trajectory. Finally, we were able to highlight the occurence of an effect induced by the use of the automation during later manual driving LC situations. This induced effect was resulting in a lowering of the decision threshold as well as the risk assessment of the LC manoeuvre to be executed. These results suggest that the human-centred design approach to driving automation systems design is promising in terms of technology acceptance and adoption. However they may present some potential induced "side" effects that will require deeper investigation.

Event localization