COSMODRIVE

The « COSMODRIVE » cognitive simulation model

Objectives:

COSMODRIVE (for COgnitive Simulation MOdel of the DRIVEr) is the fruit of a long-term research programme undertaken at LESCOT over more than 20 years (6 theses defended between 1998 and 2014, and one in progress), devoted to the cognitive modelling and computer simulation of all the processes involved in driving a vehicle. From a theoretical point of view, this simulation model is based on activity theories from ergonomics, Piagettian constructivism (e.g. driving schemas & patterns), as well as on a range of theories from the cognitive sciences (e.g. Neisser's perceptual cycle). Neisser's perceptual cycle, mental representations and situational awareness, natural decision-making), social psychology (e.g. attitudes to risk and risk-taking) or artificial intelligence, with the aim of understanding them in a 'unified' way within the framework of an 'integrated' computer simulation. The final objective of this work is to have a 'digital incarnation' of these theories in order to simulate on a machine the way in which a human driver gathers and processes information in his environment, imagines the driving situation in which he is evolving, anticipates events and assesses risks, makes decisions and then plans and implements driving actions in order to progress in the road environment.

In summary, the driving activity in COSMODRIVE is considered as a dynamic « Perception-Cognition-Action » regulation loop (see Figure 1), involving « explicit » (controlled processes and symbolic representations) and « implicit » (sub-symbolic representations and automatic processes) levels of consciousness.

Figure 1: Driving activity as a dynamic control loop

In perceptual terms, the COSMODRIVE model is based on 2 complementary processes: a bottom-up process of progressive integration of information, and a top-down process of active exploration of the road environment. COSMODRIVE is also equipped with a « virtual eye » capable of digitally simulating drivers' visual strategies, based on empirical data collected by an occulometer (Figure 2).

Figure 2: Simulation of visual strategies

The product of this perceptual processing takes the form, at the cognitive level, of mental representations (explicit or implicit; cf. Bellet et al., 2009) of the current and future driving situation. These representations are not simple copies of objective reality. They correspond to the driver's « situational awareness » of his environment. Based on empirical know-how acquired through practice (modelled in COSMODRIVE in the form of driving schemes), they form functional, simplified and distorted mental models of reality, corresponding to the way in which the driver apprehends and understands situational reality, as a function of the goals he is pursuing, the experience he possesses, the processing capacities at his disposal and the attentional resources he is mobilising at that moment for the driving task. It is on the basis of these representations that the driver assesses the risks and makes decisions to plan his actions. In COSMODRIVE, this decision-making stage is based on a « cognitive deployment » process (Figure 3).

Figure 3: Simulation of the lane change decision-making process

In addition, the challenge of this modelling work is not only to simulate « optimal » driving performance, but also to be able to account for the difficulties encountered by human drivers at the wheel, and to simulate the driving errors they may make under certain conditions (in the event of prolonged visual distraction, for example, as illustrated in Figure 4).