Helios de Rosario Martínez*, Juanma Belda Lois*, José S. Solaz Sanahuja, José Laparra Hernández, Nicolás Palomares Olivares, Elisa Signes i Pérez, Rakel Poveda Puente, Clara Solves Camallonga
Instituto de Biomecánica (IBV). Universitat Politècnica de Valencia. Edificio 9C. Camino de Vera s/n (46022) Valencia (Spain)
* IBV’s Healthcare Technology Group, CIBER in Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN)
The complexity of new vehicles, which provide the driver with a high amount of information and the ability of autonomous driving at certain moments or stretches, requires new tools to assess the impact of technology on the driver in very early stages of the development. The objective of the work being developed within the framework of the DIVEO project (project in cooperation IVACE 2016) will make it possible to develop a series of virtual drivers that will be very useful in the analysis of the effect on safety and the driving mode of the new systems that are progressively being incorporated into new car models. The driver cognitive models being developed are software tools that allow for future extensions and modifications, and which enable the analysis of systems that are at an embryonic state.
INTRODUCTION
Driving a vehicle is a complex task that involves attention mechanisms and control actions at different levels and at different time scales: drivers need to make the necessary decisions to reach their objective, depending on the traffic conditions and the environment at any given time, which involves developing sophisticated control and motor coordination mechanisms.
In addition, the increasing number of automated systems in automobiles and the pervasiveness of the information and communication elements that we continuously use (such as mobile phone terminals, navigators and other infotainment devices) make it increasingly important to know how the interaction between the driver and the various elements inside the vehicle takes place in order to ensure safe driving.
This need is even more evident in the new driving paradigms that are currently gaining momentum thanks to the technological advances in the automotive industry, such as autonomous driving by means of "smart vehicles". During certain stretches in long journeys, these vehicles give the driver a greater degree of freedom to drive and perform other secondary tasks simultaneously, although this situation may change in very short spaces of time.
For all that, the evaluation of the driving position in recent years has shifted from the purely experimental field to the analysis of cognitive and mental processes. Following this tendency and within the framework of the DIVEO project, developed in collaboration with the ITI and AIMPLAS Technological Institutes, the objective of the IBV was to develop new tools which allow researchers to simulate these complex processes for multiple typologies of scenarios and people. This will help us to define the cognitive workload management functions, with the aim of resolving potential conflicts between individual functions and their interaction with the driver. The result will significantly contribute to the design of future driving positions and their interfaces with the driver (Engström and Hollnagel, 2007).
SELECTION OF THE COGNITIVE MODEL
The work carried out in the first year of the DIVEO project is based on an update of the existing scientific and technical knowledge about the use of driving-oriented cognitive models and the computer tools available to implement them. The study performed found several approaches, although with a great disparity and an uneven proportion between developed theoretical frameworks and practical implementations. After such study, it can be stated that, despite the large number of theoretical cognitive models related to various aspects of driving, there is no accepted common framework for driver behavior modeling.
Among the great variety of strategies to model the cognitive processes of the driver that were identified in the bibliography, the development being carried out in the DIVEO project focuses on the category of functional models of information processing, in which the cognitive activity is characterized as a sequence of computational steps, including perception, decision making and selection of the most appropriate response (Engström and Hollnagel, 2007). The order and times associated with the execution of these sequences depend on the mechanisms of attention and the modeled limitations of the driver cognitive resources. In the case of vehicle driving, there are models of this type that have been applied to more general driving behavior models, such as Wickens' model (2002), which considers that the cognitive resources that a person dedicates to a task are structured into four "dimensions": the dimension related to the type of information used (spatial or verbal), the means by which this information is transmitted (auditory or visual), the way in which the interaction with the environment takes place (perception of stimuli, body response), and—if it is a task of perception—how the information to be processed is distributed (focused on a point or spread over the environment).
Figure 1
According to this model, the capacity of human beings to perform complex tasks or to simultaneously perform different tasks depends not only on the amount of information to be processed and the time demanded (including aspects such as the urgency or priority assigned to the tasks), but also on the extent to which the various "subtasks" can be divided between the different types of cognitive resources or if they compete for them.
Operationally, this results into matrices with "concurrence costs", which allow us to quantify in the multiple dimensions the cognitive demand of one or several tasks, such as driving looking for a destination, and at the same time paying attention to traffic signs, other vehicles and road events, in-vehicle information systems, etc. On the basis of these quantities, and considering limited cognitive resources, the speed of response to stimuli (or failure to respond if the tasks have a limited response time, such as avoiding an obstacle) can be determined.
Figure 2. The driver will receive more and more information and stimuli. The impact of technology on mental burden is critical
From the point of view of implementation, the various types of driver cognitive models have been studied: those based on cognitive architectures such as ACT-R and SOAR (Salvucci, 2006), and those based on series of discrete events, such as IMPRINT, an application of the United States public administration originally aimed at modeling the driving of military vehicles.
Cognitive architectures are a general framework in which modular models of the mind are defined, with an approach close to the developments of Artificial Intelligence (AI), in order to represent human cognitive capacities in a computer system. Although they have a great potential to accurately represent how the mind works and can model the solution of complex problems, their programming is very complex. On the other hand, the models of discrete events are not specifically intended to model the mind, but to more general scenarios in a simplified way, such as a set of processes and events that interact with each other, causing their triggering, ending or interruption, depending on how the sequence of events develops. For example, in a driving scenario, processes and events can be defined, such as the distance traveled by a vehicle, the encounter of the vehicle with a sign, obstacle or other elements of the environment, actions such as accelerating, braking, changing directions, actions through which drivers interact with what they perceive from the environment and the vehicle controls, etc.
Models based on discrete events are more affordable for scenarios with low or medium complexity and enable a more direct integration of rules that relate stimuli to responses, which is one of the main objectives when assessing the interaction of the driver with the new interfaces of the vehicle, rather than understanding how the mind works. For these reasons, it was decided to implement this type of discrete event models, using the cognitive workload rules defined in Wickens model.
IMPLEMENTATION OF THE MODEL
As a practical application of these cognitive models, it was decided to implement a simulation scenario of the Lane Change Test: a validated and widely used methodology to assess the cognitive workload and the safety of systems that can distract the driver attention by means of a driving simulator. At the conceptual level, it is a simple test that involves traveling a distance and simultaneously perform secondary tasks (turning on the radio, paying attention to the directions of the navigation system), while signs to change between three lanes appear on the road. The reaction capacity and the path traveled by the vehicle are evaluated.
Figure 3. Tests in the IBV driving simulator
The implementation of this method is standardized through ISO 20622 (2010). It is one of the tools that the IBV has set up in its driving laboratory and is used in the assessment of products and on-board systems that can be a potential distraction for the driver. The aim of the DIVEO project is to extend the application of the standard from a prototype assessment tool to a design assessment tool by using "virtual" users.
One of the ambitions of this objective is to use open and widely used computer tools that support the independence of specific and proprietary solutions in order to facilitate transferability and extension. With this in mind, the Python programming language was chosen: a free, public and multipurpose language that allows for implementations, offers very different uses for research, industrial and commercial purposes, and has a large community of users and developers. One of the developments is the SimPy "library" for discrete event simulation, potentially useful for simulating vehicle paths in various types of scenarios. During the first year of the project, the work focused on an extension of the "environments", on a study to verify the feasibility of using SimPy to create scenarios related to the control of a vehicle with increasing complexity. Several scenarios were programed, from the simplest and most elementary with a single vehicle traveling a path with no interruptions, to one in which several drivers interact with obstacles in a section of road, using models based on safety margins, so as to try to avoid them, taking into account different reaction capacities of both the driver and the vehicle itself, according to the vehicle speeds and the distances at which the driver is able to detect obstacles.
Thus, virtual cars, road sections and obstacles were created, which can interact with each other. Environments where these elements coexist and interact were also created, as well as a process that makes it possible to monitor (observe) the situation.
Although events (relevant events) are dynamically defined, there must be monitoring processes at time intervals for the model to know when an event must be generated. However, using fixed intervals is not effective, because they should be as small as the minimum reaction time or environment change time, whereas during most of the path no relevant events or changes are expected (cars move along the road without anything special happening). To solve this, it has been necessary to program variable waiting times that automatically consider when a change in the interactions may occur, due to future events.
For example, if the driving of a vehicle is in a "stationary" state, at a constant speed, time jumps forward for that vehicle until there is an element that alters that state, including signs, obstacles or other objects within the visual range of the driver, that may alter the driving. During the maneuvers performed in that interaction, a minimum interval is considered. What several vehicles do can be simulated in parallel at different intervals, as long as they are separated enough to avoid interaction between them. This makes it possible to carry out complex evaluations in a short period of time, since we eliminate from the calculation the time during which nothing relevant happens.
CONCLUSION
The SimPy architecture has proven to be suitable to simulate the proposed scenarios of interaction between vehicles, roads and obstacles. The amount of code required increases progressively as the quantity of elements and interactions becomes more complex, although this increase in the code seems to be controllable.
The next step is to model the cognitive processes. To do this, the objects implemented in SimPy can be used, associated with a new object type: "driver". Objects related to the "macro-scale" of the road (vehicles, obstacles, lanes, signs, etc.) can be added to the model developed, without interacting with the cognitive models. Furthermore, the driver cognitive model should only act with vehicle control (speed, acceleration, braking ...), as well as with some external elements that may enter the visual range in perception processes.
REFERENCES
Engström, Johan, and Erik Hollnagel. 2007. “A General Conceptual Framework for Modelling Behavioural Effects of Driver Support Functions.” In Modelling Driver Behaviour in Automotive Environments, 61–84. Springer. http://link.springer.com/chapter/10.1007/978-1-84628-618-6_4.
Kandemir, Cansu, Holly A. H. Handley, and Deborah Thompson. 2016. “A Workload Model to Evaluate Distracters and Driver’s Aids.” International Journal of Industrial Ergonomics. Accessed September 22. doi:10.1016/j.ergon.2016.09.004.
Salvucci, Dario D. 2006. “Modeling Driver Behavior in a Cognitive Architecture.” Human Factors: The Journal of the Human Factors and Ergonomics Society 48 (2): 362–80. http://hfs.sagepub.com/content/48/2/362.short.
Wickens, Christopher D. 2002. “Multiple Resources and Performance Prediction.” Theoretical Issues in Ergonomics Science 3 (2): 159–77. http://www.tandfonline.com/doi/abs/10.1080/14639220210123806.
