Information Technology Reference
In-Depth Information
- The
resolution
phase (begin overtaking, in the case of Figure 5.2) ends in the
disappearance of the obstacle (it has been overtaken, or has turned off the road, for
example) and the return to the initial driving conditions (“driving without an
obstacle”).
These different activity phases are not systematically required (overtaking can
occur as early as the phase of obstacle detection) and do not always proceed in a
sequential logic. It is more a matter of a “regulation cycle” (modeled here in the
form of an automaton of defined states) composed of an ensemble of potential states
enabling us to move from one activity phase to another, according to certain
transition conditions.
These transition rules can be based on behavioral indicators (the fact that the
driver presses the brake, for example), as much as on variations of situational
parameters (for instance the distance to the obstacle). Beyond the identification of
the
activity phases
in real time,
the objective of the model was to judge how
adequate the implemented behaviors were with regards to the criticality of the
situation (estimated from the TTC), thus leading to an “integrated” criticality
diagnosis (see Figure 5.1). The stake was then to estimate whether or not these
different activity phases were mastered by the driver, and to specify the nature of the
assistance required in real time (aid for detection of the obstacle, decision making,
carrying out an action, etc.).
The results obtained at the end of the project [BEL 06] have enabled us to show
the benefit of this approach from an ergonomic point of view, and to prove its
scientific and technical feasibility (80% good diagnosis in the situation of
approaching a fixed or slow vehicle), as soon as the critical events were correctly
detected by the perception technologies.
Though the developed prototype is still far from being operational, this research
project (above all focused on the diagnosis stakes) shows us how to resort to models
for the analysis of human activity in real time to enable adaptive management of
human-machine interactions (Does the driver need assistance? What does the aid
provide? In what form?). It nonetheless has not enabled us to explore the effective
implementation and comparative evaluation of different modalities of human-
machine cooperation in a driving situation. This was possible, however, in the
monitoring research case.
