=Paper=
{{Paper
|id=None
|storemode=property
|title=The design of preventive safety systems: a cognitive engineering problem
|pdfUrl=https://ceur-ws.org/Vol-696/paper5.pdf
|volume=Vol-696
}}
==The design of preventive safety systems: a cognitive engineering problem==
https://ceur-ws.org/Vol-696/paper5.pdf
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
The Design of Preventive Safety Systems: a Cognitive Engineering Problem
Caterina Calefato (calefato@re-lab.it)
University of Turin Department of Computer Science Corso Svizzera 185,
10149, Torino, Italy
Luca Minin (luca.minin@unimore.it), Roberto Montanari (roberto.montanari@unimore.it)
University of Modena e Reggio Emilia, HMI Group DISMI, Via Amendola 2,
42100, Reggio Emilia, Italy
Abstract
The precondition analysis
In this paper a design framework for preventive safety
Accidents occur because multiple factors combine to
systems (ADAS) is proposed. The design framework takes
into account risk mitigation strategies, advanced driver’s create the necessary conditions for them. Over the past 30
model, based on modern approaches and algorithms years, the literature shows a consistent trend in trying to
(machine learning and add-on functionalities), able to understand accidents in aviation, nuclear power
capture key aspects of human behavior, such as distraction, generation, telecommunications, unmanned and manned
and to retain the fundamental characteristics of cognition spaceflight, railroad transport, shipping, healthcare and
and decision making. many other fields (Catino 2002). Regardless of the
domain of investigation, there are some crucial questions
Introduction to which the research is trying to find an answer (Cook &
Driver modeling is a scientific area involving several O'Connor 2005):
disciplines, such as psychology, physics, computer • How do accidents happen and what do they
science, etc. The importance of adding the learning mean?
capability to information systems, in order to make them • Are accidents foreseeable?
more effective and smarter, is confirmed by the variety of • If so, are they preventable?
areas in which user’s modeling has already been applied: • What role does technology play in accidents?
information retrieval, filtering and extraction systems, • What role does human performance play?
adaptive user interfaces, educational software, etc. • Are accidents evidence of systemic problems
In relation to the problem formulated above, the aim of or are they isolated failures?
this paper is to deeply understand the problem of ADAS • If accidents are systemic, how can the system
(Advanced Driver Assistance System Design) design, the be fixed to prevent future accidents?
problem of developing an effective driver’s and driving What it may be interesting to study is the detection of an
model supporting distraction mitigation. Such systems accident pre-conditions and which may be the conditions
would mitigate the effects of distraction and tolerate the combination (circumstances) that may lead toward an
consequences of distraction thanks to a better road and accident. For each combination (that we may label risk
vehicle design (Regan, Lee & Young 2009). layouts) a mitigation strategy will be applied in order to
A feasible and promising solution is the use of Add-On avoid a possible accident. “What was lacking was the
functionalities, able to detect driving maneuvers that are ability to foresee that circumstances would conspire to
indication of distraction, placing them in the framework create the conditions needed to make these technical
of a cognitive model of human behavior. features active and lethal” (Cook & O'Connor 2005).
In this paper a design framework for preventive safety An adaptive system should be able to detect risk
systems is proposed, following three main building layouts and dynamically adapt its behavior in order to
blocks: avoid accident, but the failure factor in this scheme is
1. new knowledge about driver behavior: change. In a complex system formed by a context, a
extensive empirical studies about the sources of predictable system (including automatic applications) and
accidents and potential counter measures as a the human being, the unpredictable factor is the human
basis for the driver model development. being behavior and its combination to a certain context.
2. risk mitigation strategies: implementation of a Although information technology can defend against
human error risk based approach; some types of accidents and failures, the impact of
3. advanced driver modeling: development of automation on human-machine system performance is a
models for predicting correct and erroneous mixture of desirable and undesirable effects (Perry,
driver behavior, based on modern approaches Wears & Cook 2005).
and algorithms (machine learning), able to Systems like ADAS (Advanced Driver Assistance
capture key aspects of human behavior, and to Systems) have great potential for enhancing road safety,
retain the fundamental characteristics of but on the other hand the safety benefits of ADAS may
cognition and decision making. be significantly reduced by unexpected behavioral
responses to the technologies, e.g. system over-reliance,
safety margin compensation and distraction, leading
34
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
toward an automation failure. The automation failure is a vehicle timely providing vehicles’ stops and
side effect of an effort to produce ‘‘safety’’(Catino 2002). small movements.
• Lane Keeping Assistant: The function of a
ADAS: existing applications lane keeping assistant system includes the
In addition to the safety issues associated with the driving lane detection and the feedback to the driver if
task, the proliferation of complex in-vehicle functions he/she is leaving a defined trajectory within
itself poses a further challenge for the design of the the lane. An active steering wheel can help the
driver-vehicle interface: one of the current research area driver with a force feedback to keep on this
in automotive is the development of preventive warning trajectory. The lane is detected by a video
systems, also called ADAS (i.e. Advanced Driver image processing system.
Assistance Systems) adopted with the aim of improving • Local Hazard Warning : If a hazard occurs
driving safety. These systems are able to detect an far away in front of the vehicle, so that the
incoming dangerous in advance, allowing a time to driver cannot see it, this system will warn
perform a repairing manoeuvre. him/her. By the means of communication it is
ADAS are aimed at “partly supporting and/or taking possible, to transfer this information over long
over the driver’s tasks” (Berghout, Versteegt, van Arem distances
2003) so to generally provide safer driving conditions. • Lane Change Assistant: Before and during a
Several functions can be mentioned within ADAS set. In dangerous lane change process, the lane
the following, a list of the main relevant ones is reported change assistant will warn the driver. Several
(Ehmanns & Spannheimer 2004): stages of such a system are possible from pure
• Lane departure warning: If certain warning systems to even haptic feedback at
thresholds (like distance, time to lane the steering wheel to help the driver following
crossing) allow a prediction of a lane a lane change trajectory.
departure this system warns the driver by • Blind Sport Monitoring: This function
means of acoustic, optic or haptic feedback. detects if a vehicle is present in the so called
The detection of the lane markings results “blind spot” area when the vehicle is starting a
from e.g. video image processing. In order to lane change and/or overtaking maneuvers. A
have a robust lane marking detection two camera is placed into the left rear-mirror and
needs can be absolved: (i) good visible lane once the incoming vehicle is recognized, a
markings have to be provided by the warning is issued to the driver.
infrastructure and (ii) a robust lane detection • Obstacle & Collision Warning: The driver
sensing system has to be implemented in the will be warned if a potential collision is
vehicles. Both aspects are influencing the detected via radar-based technology (e.g.
complexity of the system on the roadside and another car or obstacle). The functional limits
the technical level. of these systems have to be clearly pointed
• Near field collision warning: The near field out. The liability problem of these systems
collision warning includes the detection of grows with the complexity of the detecting
especially vehicles in the near field like in the scenarios.
blind spot area. The detection area is very • Obstacle and Collision Avoidance: This
close limited to the vehicle. Suitable sensor system has an extended functionality
systems for the detection of other cars are compared to the Obstacle and Collision
radar or vision based sensors. Warning. An autonomous intervention takes
• Curve & speed limit info: These systems over the control of the vehicle in critical
inform the driver about speed limits and the situations in order to avoid an accident.
recommended speed in curves. Possibly the Longitudinal and lateral control will be done
necessary information can be taken from by the system during the defined time while
digital maps, image processing the dangerous event takes place.
communication systems between vehicles and • Night Vision: Based on camera techniques
infrastructure. like near or far infrared, it allows enhancing
• Adaptive Cruise Control (ACC) /Stop & the perception of the driver in dark light
Go: The ACC and Stop & Go establish a conditions. The picture of the camera will be
virtual link with the frontal vehicle via a shown to the driver by monitors or head up
radar-based technology and keep booth displays.
vehicle within a safe distance. The main • Platooning: Several cars are connected
innovation of this systems, that is derived electronically (e.g. by the means of
from the well-known cruise-control, is that the communication) and follow one after the other
distance can be adapted both to the driver’s in a platoon. An example is the connection of
preferences (as in ACC) and to the specific trucks in order to save space, fuel and to
requirements of the urban environment (as in increase the traffic flow. As the following
the Stop & Go). In traffic condition as in a vehicles are driven automatically, the system
queue, the Stop & Go automatically drive the is complex concerning all aspects. The
35
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
takeover of the driver at e.g. gateways has to Risk mitigation strategies: recommending the
be taken into account as well as the behavior accident avoiding actions
in mixed traffic at driveways.
ADAS can be considered an application of
recommending systems that recommends the driver
Designing the trustiness: ADAS research issues
repairing maneuvers in order to avoid an accident. The
Interaction with these devices is one of the many most advanced systems are able to directly take part in
activities that constitutes driving and so it can represent the driving task, whether the driver doesn’t react on time.
an additional source of driving-related distraction (Regan, Also in this case, the systems follows a recommendation
Lee & Young 2009). For example poorly designed formulated by the system itself.
collision warning systems may be even more likely to Recommender systems have become a promising
distract drivers; navigation represents a driving-related research area since the appearance of the first papers on
task with substantial potential to distract (Neale et al collaborative filtering in the mid 1990s (Hill et al1995),
2005), (Dingus et al 1989). (Resnick et al 1994), (Shardanand & Maes 1995)
The analysis of ADAS working conditions, Shortly, the recommendation problem is formally
architectures and performances leads towards the represented as a space S of possible items that may be
definition of a proper theoretical framework that is not very big, ranging in hundreds of thousands or even
yet present in current projects. millions of items in some applications, such as
The reasoning behind is the following: Advanced recommending books. An utility function measures the
Driver Assistance Systems, or ADAS, are systems that usefulness of each item for a certain user.
help the driver in its driving process: they detect a In recommender systems, the utility of an item is
dangerous situation and gives a warning. We can define usually represented by a rating, which indicates how a
analytical (Andreone et al 2005), the ADAS type that particular user likes a particular item or how a particular
warns the driver suggesting accident-avoiding item is appropriate for a certain user, taking care of a set
maneuvers. The ADAS is behavioral if it acts in place of of context conditions. Generally speaking , utility can be
the driver, partially taking over a certain driving task an arbitrary function, including a profit function.
(Andreone et al 2005) (Hoch et al 2007)0. Depending on the application, the utility can either be
In the case of analytical ADAS we can consider there specified by the user, as is often done for the user-defined
are two actors playing a role: ratings, or is computed by the application, as can be the
• The driver case for a profit-based utility function (Adomavicius,
• The warning system Tuzhilin 2005).
In the case of behavioral ADAS we can consider there To each element of the user space C can be associated
are three actors playing a role: a profile that includes the user characteristics that are
• The driver relevant for the current application. Similarly each
• The horse (the artificial system able to drive element of the item space S is defined by a set of relevant
in place of the driver, see Flemisch et al 2003) characteristics.
• The warning system In recommender systems, utility is typically
In both cases, analytical and behavioral ADAS, there is represented by ratings, therefore, the recommendation
a warning systems that detects the dangerous situation engine should be able to estimate (predict) the ratings of
and then provides the driver, as safety warning, accident- the nonrated user/item combinations and issue
avoiding maneuvers. appropriate recommendations based on these predictions.
The purpose of these systems is to foresee and detect Extrapolations from known to unknown ratings are
possible driver’s errors and mistakes, due to a usually done by:
misbehavior such as distraction, or resulting from too • specifying heuristics that define the utility function
high workload, missing perception, wrong and empirically validating its performance
action/execution or poor operator skills. • estimating the utility function that optimizes certain
The ADAS design is aimed at enhancing the driver’s performance criterion, such as the mean square error.
perception of hazards and critical situations (in some
cases, by partly automating the driving task as well). Of Despite of the results nowadays achieved, the existing
course the potential of such systems in reducing accidents generation of recommender systems still requires further
depends on the effectiveness of their interaction with the improvements including better methods for representing
driver. For example, in the case of an anti-collision user behavior and the information about the items to be
systems it is safety-critical that the collision warning is recommended, more advanced recommendation
able to generate the appropriate feedback (e.g. an modeling methods, incorporation of various contextual
avoidance maneuver). information into the recommendation process, utilization
Since ADAS can be actually considered recommending of multicriteria ratings, development of less intrusive and
systems, the use of an appropriate driver’s or/and driving more flexible recommendation methods that also rely on
model will improve their effectiveness and consequently, the measures that more effectively determine
human safety. performance of recommender systems (Adomavicius,
Tuzhilin 2005).
In the case of services provided on board a car, we can
notice that they are rapidly growing. Almost all car
36
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
manufactures are offering systems that add functionality In-Vehicle information system (IVIS) is activated and the
to route planners, possibly integrated with internet and driver is asked to interact with it, the driver him/herself is
web access or that support driver in high demanding distracted from the driving task, that is, his/her attention
tasks, in order to increase safety and avoiding accidents. is moved from the driving task to the secondary task. A
The availability of these add-ons is an interesting relevant part of vehicle crashes are estimated to
opportunity, considering that nowadays the amount of The Driver Assistance Systems have to be able to adapt
time spent in the car (e.g., for commuting or for work and their action to the context and to the driver and vehicle
vacation trips) is very high (Console et al 2003). status. Thereby, they need a model of human behaviour
If on one hand the driver and the other vehicle that takes into account the model of the system
occupants can actively use the time spent on the car, on performance and that is able to detect and classify
the other hand the use of these services can be distracting driver’s intention and distraction, in order is essential to
and can create serious safety problems (Green 200) facilitate operating mode transition between users and
(Console et al 2003), contracting societal goals of driver assistance systems.
increasing safety, reducing the number of accidents. As a The need of an effective user model is a requirement for
consequence it is necessary to find a proper compromise any recommending system, as faced and confirmed by
between the increasing number and complexity of the the domain literature on user modeling and automatic
services and the need of making the services compatible recommending systems. This requirements is crucial for
with the fact the user is driving. any recommending system that has to cope with time-
Starting from this consideration, the introduction of criticality, that directly affects safety.
personalization and adaptation strategies and techniques ADAS applications are examples of such systems and
should be a feasible solution in the case of services in the they represent a challenging test bed for the
car. In fact, by considering the characteristics of the user implementation and validation of user behavioral
and the context of interaction, a personalized and modeling systems realized by means of Machine
adaptive system may tailor the interaction to the way Learning techniques.
which is most appropriate to avoid distractions, and as a Basically, the human behavior is characterized by the
direct consequence, to avoid an accident (Console et al interactions between driver-vehicle and driver
2003). environment.
In the case of safety-critical systems which should The first interaction is related to how the driver
recommend accident avoiding maneuvers, the adaptation interacts with the vehicle and all systems and sub-
of the recommendations to the specific user is crucial, systems on-board.
according to the psychophysical parameters that are taken The second interaction is related to how drivers
into account (i.e. mental workload, distraction, arousal perceive and process the data coming out from the
level, situation awareness). In the case of advanced surrounding scenario.
driving assistance systems one of the most important Hence, the driver model should be adaptive to different
psychophysical parameter to be taken into account is drivers’ style and preferences as well as to the external
distraction. The system should be able to assess driver’s environment (including learning both from the driving
distraction in order to estimate accident precondition (risk experience and from the surrounding conditions), but
layout) and recommend driver appropriate actions, or in overall it should allow to assess and foresee distraction.
the case of adaptive automatic systems to perform a Preventing distraction permit to prevent driving errors
proper risk mitigation strategy. and accident risk, as a consequence, a risk based design
If the recommending engine has not at its disposal a approach (that follows a risk mitigation strategy) is
user behavior model, it can formulate recommendation crucial for the design of vehicles and transport systems in
that may lead towards no decisions or wrong decisions. order to guarantee safety and efficiency of human
Whether the system prediction capability is augmented mobility.
through a user behavior model it is possible to reduce User modeling (UM) aims at improving system
errors and then the risk of accidents. This consideration is effectiveness and reliability by adapting the behavior of
of paramount importance in complex safety critical the system to the needs of the individual.
systems as avionic and automotive, that commonly use The importance of adding this capability to information
different kind of recommending services. systems is proven by the variety of areas in which user
modeling has already been applied: information retrieval,
The design of cognitive preventive safety filtering and extraction systems, adaptive user interfaces,
systems educational software, safety-critical systems .
Machine learning (ML) techniques have been applied
Driving is considered as a complex and multitasking
to user modeling problems for acquiring models of
cognitive activity that can be summarized by four main
individual users interacting with an information system
sub-processes: perception, analysis, decision and action.
and grouping them into communities or stereotypes with
To be performed, each phase presumes the achievement
common interests. This functionality is essential in order
of the previous one. That said, it is likely that the
to have a useful and usable system that can modify its
demands of one element of driving will interfere with
behavior over time and for different users (Langley
another element.
1999). As elicited from literature (Tango Botta 2009),
ADAS new technologies have great potential for
(Tango et al 2009) (Tango et al 2010) there is a trend in
enhancing road safety, however, when an ADAS or an
choosing machine learning techniques in the study of
37
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
modeling of human behaviors, that is non-deterministic installation of further in-vehicle sensors and devices
and highly non-linear. (Lolli et al. 2009) .
Driver Assistance Systems have to handle crucial An alternative is the use of the so-called Add-On
aspects like timing and warning, therefore the Functionalities (AOF). They do not require new sensors
development of an algorithm for the personalization of but only information coming from the on-board network
such aspects that takes into account for example is and sub-systems (e.g. the elements of chassis,
needed. suspensions, steering angle, etc.). This information is
Regarding the drivers’ intention prediction, several computed in a well tuned algorithm and the results
models have been proposed aiming at reproducing in a provide some added-value supports to the drivers as
virtual environment how the drivers could behave AOFs for pre-crash detection. They prepare the vehicle to
according to specific Driver, Vehicle or Environment the impact in critical situation which can not be avoided,
conditions , that is DVE model (Tango et al. 2010) . In for example, by pre-tensioning the seat belt (Lolli et al.
the domain literature there are different approaches like: 2009). They can be used also to indirectly infer form the
• the IPS (Information Processing System), driving & driver data crucial information about driver’s
which has been applied in almost all behavior like distraction, workload and arousal.
technological fields to describe human Add-On Functionalities can in fact be divided in two
interaction with control systems, at different main categories:
levels of automation (Neisser 1967) • Driving Behaviour: i.e. Add-On Functionalities
• the PIPE (Perception, Interpretation, Planning related to driving performance. Main objective
and finally Execution) based on a very simple of these AOF is to estimate driving conditions
approach that assumes that behaviour derives concerning road, dynamics and current
from a cyclical sequence of four cognitive manoeuvre.
functions in brackets. (Cacciabue 1998) • Driver Behaviour: these AOF deal with driver
The development of a model of the human machine current state, mainly intended as mental effort
system is driven by the model of the Driver, which is the (or workload) related to the driving task.
most complex element of the system.
Concerning the design of algorithms used to represent In order to be implemented, an Add-On function has to
the Driver behaviour, previous and ongoing studies satisfy two conditions:
propose different approaches based on the real-time • All Add-On inputs must be available and
monitoring of the drivers’ performance (Lolli et al 2009) shareable.
(e.g. variation of the position on the road, speed, steering • At least one Add-On output can be received as
wheel movements) or the drivers’ physiological status input by a vehicle device.
performing primary and secondary tasks (e.g. eye gaze, An Add-On Function with m inputs and n outputs is
eye movements, heart frequency variation, galvanic skin defined as:
response, etc.) (Ji & Yang 2002).0 , , … , , …
Basing on these information, these approaches allow to or in a concise form:
predict specific drivers’ profiles (e.g. stressed,
aggressive, tired, distracted, high workload etc.) and are If Y is set from available device output and X is set
developed following machine learning approaches. from available device inputs, the two aforementioned
Thanks to machine learning, information on drivers’ conditions can be expressed as follows:
profile can be automatically extracted from data, by In order to be implementable, an add-on functionality ,
computational and statistical methods applied to defined as
observable information (e.g. drivers’ performance data). ,
On one hand the assessment of the driver’s status and
with , , … and , , …
consequentially the prediction of his/her next behaviour
must satisfy the following conditions:
is easier and most successful using driver’s physiological
1. , , …
data, as for example the eye gaze and the eye movements
2. , 1, … , |
measured by means of eye tracker. On the other hand
where and are respectively set of devices inputs and
eye-tracking is an intrusive measurement system of
outputs.
distraction, it represents a further equipment and a further
cost that stand in the way of a next future mass-
In the study presented in (Lolli et al. 2009) the efforts
marketing. What is really interesting and challenging is to
are focused on the use of AOF in order to collect useful
obtain a driver index analysing driving performance data,
data that will be employed to fill in the driver’s profile
realising what it may be considered an ADD-On
and status and to log his/her performance. All this
Functionalities.
information will be used as a trigger for the adaptive
Understanding driver’s maneuvers by the use of automation applied to the in-vehicle information systems
(IVIS) or to the Advanced Driver Assistance Systems
Add-On Functionalities (ADAS). The development of AOFs has been tested for
Research in driver comfort and performance tuning in a simulated environment using data
improvement understanding driver’s maneuvers is very Matlab/Simulink vehicle model. Identified AOF outputs
active. Usually, these targets are achieved through the
38
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
to improve driver performance and safety are the Speed Deceleration Jerks TC
following: (DJ)
Driver Stress (DS). From previous studies (Gulian et al. Brake Pressure Braking Frequency TC
1989), stress can be related to lateral vehicle control. (BF)
Then, it could be derived by the steering activity: several Accelerator Accelerator TC
indexes for lateral control monitoring were provided, in Displacement Frequency (AF)
particular: HFS (High Frequency Steering) (McDonald Gear number Gear Index (GI) TC
Hoffman 1980), RR (Reversal Rate) (McLean Hoffman Z acceleration Frontal Obstacle RC
1975) and SAR (Steering Action Rate) (Verwey 1991). Preview (FR)
According to the level of stress it is possible to find Roll rate Roll Index (RI) RC
strategies to assist the driver in particularly demanding Pitch rate Pitch Index (PI) RC
maneuvers by modifying the steering force feedback, Suspensions Frontal Obstacle RC
braking behavior and inhibition of secondary task to Displacement Preview (FR)
avoid possible safety-risk situations. Inputs were selected to compute specific AOF
Particularly, modifying the steering force feedback or parameters with the aim to describe driver stress, traffic
the braking behavior has an interesting impact in the congestion and road conditions. AOF outputs are the
human-machine interaction, because the feedback to the result of the balanced sum among parameters; for
driver is haptic Empirical test confirmed that driving instance, the Driver Stress (DS) index (2) was developed
performance significantly improved when the system as follow 0:
activated the force feedback models. DS = (RR x cRR) + (HFS x cHFS) + (SAR x cSAR) (2)
These results compared with data arose from other Where cRR + cHFS + cSAR = 1 are the coefficients to be
studies in the literature (Steel Gillespie 2001) suggested tuned in order to define the final value of the AOF
that, using an intelligent, haptic steering wheel rather output. Each AOF outputs (TC, DS and RC) and their
than a traditional passive steering wheel, drivers are related computed parameters (see “AOF parameters” in
better able to closely follow a reference path while Table 1) were developed in a simulated environment
requiring fewer visual cues (Minin et al. 2009) using Matlab/Simulink (www.mathworks.com).
For a decade researchers (Bertollini Hogan 1999) have In order to test and tune these parameters, AOF models
been finding that the presence of haptic feedback on the were interfaced with a Matlab/Simulink simulated
steering wheel could help drivers to perform a visually- vehicle. The whole model (AOF and simulated vehicle) is
guided task by providing relevant information like fed by real driving data coming from a professional
vehicle speed and trajectory. Referring to the augmented driving simulator. Specific tests were carried out, aiming
cognition field, we can assess that when using a haptic to provide driving situation where each parameter varies
assist steering wheel rather than a traditional passive significantly; then, their effectiveness was assessed.
steering wheel, drivers are better able to follow a According to the result, information monitored by AOF
reference path and at the same time, they required fewer outputs (DS, TC and RC) will be used as a basis for the
visual cues (Griffiths Gillespie 2004). development of strategies aiming at improve driving
Traffic congestion (TC). Through the monitoring of performance, safety and comfort.
longitudinal vehicle parameters (e.g. brakes and speed AOF test and tuning: Driver Stress
behavior) it is possible to state whether drivers are In the following, test and tune of Driver Stress (DS)
driving in a heavy traffic situation or not. In this case, parameters are described. The DS index is the balanced
strategies could be elaborated by optimizing both the sum of steering angle based parameters, in particular:
engine-fuel management at a low speed and the driver SAR (Steering Action Rate), HFS (High Frequency
comfort aiming at reduces the level of stress (Lolli et al. Steering), RR (Reversal Rate). A default tuning of
2009). coefficients related to these parameters has been applied
Road Conditions (RC). The knowledge of road profile (CRR = 0,4; CHFS = 0,2; CSAR = 0,4). The effect of the
characteristics through information coming from specific tuning was assessed by comparing the expected stress
sensors (i.e. suspensions, Roll Rate Sensor, Pitch Rate profile in certain pre-determined conditions (i.e., the
Sensor, Sound Sensor Cluster, ESP intervention) allows points numbered from 1 to 5 in Figure 1) with the
alerting active suspension system to smooth the impact of drivers’ steering activity (Lolli et al. 2009).
obstacles, helping drivers to reduce the effect of this Two tests were conducted on a driving simulator where
critical situation (Lolli et al. 2009). 12 subjects, each of them was asked to drive for 10
minutes. Test environments were characterized by roads
The table below (Lolli et al. 2009) reports inputs coming with different curve radius, variable visibility (from 100
from the vehicle chassis and used to compute AOFs. to 4500 m) reproduced with fog and variable traffic (from
10 to 50 vehicle/km) (Lolli et al. 2009).
Table 1 AOF inputs, parameters and outputs (Lolli et
al. 2009).
AOF inputs from AOF parameters AOF
chassis outputs
Steering Angle HFS, SAR, RR DS
39
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
Conducted simulations are just preliminary tests to find
out the most promising indexes. It is very important to
improve techniques aimed at monitoring of the status and
the performance of both the driver and the vehicle in
order to gather all data useful to customize the
information provision strategies of the in-vehicle devices.
In this way the vehicle may adapt his status, improving
comfort or modifying driving performances.
This proactive behaviour paves the way to mechanisms
able to infer the driver’s distraction and situation
Figure 1 Driving scenario (track) (Lolli et al. 2009). awareness, allowing the triggering of adaptive
automation strategies. The information provided by
In order to increase the steering activity, subjects were AOFs are real-time and allows at dynamically
also expected to complete a secondary visual task, which implementing such adaptive strategies and referring them
consisted in pressing the left/right side of a touch-screen both to the path changes and to the driver’s status. The
on the left side of the vehicle cabin, according to the AOFs added-value is the prospect to obtain information
position of a circle displayed among smaller ones. Data in a not intrusive way (Lolli et al. 2009).
regarding DS parameters were collected and compared
with the steering activity in a specific point of the road The framework for the design of preventive
(the number from 1…5 circled in Figure 2) (Lolli et al. safety systems
2009). The aim of preventive safety system is to support drivers,
Due to the large amount of information, the especially in risky and critical situations, or whenever
comparison focused on a sub-sample of 3 subjects; mean distraction may occur. The first step for the design of a
values of steering activity and DS parameters are then vehicle able to assess the driver status and intentions is
depicted. An example of steering angle activity is showed the development a model able to explain and reproduce
in the top side of Figure 2, while Driver Stress index in driver’s characteristics. Based on the empirical results
the bottom. Both are related to scenario coordinates (x- presented in the previous chapters, this research work
axis). As foreseen, an increased steering activity leads to aims at developing a little missing piece of the puzzle of
higher Driving Stress values 0. These peaks are pointed the future intelligent vehicles: namely to identify the
out in particularly critical situations (due to curves, high main elements for a feasible architecture of a “cognitive
traffic, low visibility), highlighted in the circled number driving assistance system” which will substantially
of the figures. advance both integrated safety/assistance systems and the
cooperation between human beings and highly automated
vehicles.
The feasibility of such an architecture has been
investigated analyzing:
• The problem of accident precondition analysis
• ADAS existing applications and research issues
• Risk mitigation strategies for the accident
avoiding
• Design issues of cognitive preventive safety
systems
• The understanding of driver behavior from
Figure 2 DS compared with steering angle (Lolli et al. driving maneuvers by means of add-on
2009) functionalities
Results show that the first DS parameters’ tuning
produced an index able to detect driver stress status. Hence the framework to develop an effective model of
Since the analysis was carried out on a small sample of driver’s perception will be include four major functional
subjects, in order to increase the significance of the areas:
tuning the above deployed comparison will be extended 1. The core application, where motion-planning
to all subjects (Lolli et al. 2009). tools including enhanced personalisation, will be
Driving Stress Index appears to be a good starting used to explore the maneuver space and to
point for developing a parameter able to detect driver ultimately understand the driver, and if needed,
status in real-time even if a deeper test and tuning activity to produce maneuvers compatible to human
is needed. Furthermore, together with the other AOFs motion.
(Traffic Congestion and Road Conditions) can be easily 2. Improved sensing of driver input, where the
implemented on a vehicle ECU (Electronic Control Unit) control input (“input” in control theory sense)
or a DSP (Digital Signal Processor) (Lolli et al. 2009). produced by the driver, both in longitudinal and
lateral directions will be measured. The scope is
40
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
to better discriminate between different motion Journal of Intelligent Information Systems 21(2)
alternatives. The representation of driver input (2003), 249–284.
will be given in abstract, vehicle independent Cook R.I., O'Connor M.F. (2005). “Thinking about
way; preferably in terms of the longitudinal and accidents and systems”. In: Manasse H.R., Thompson
lateral jerk (Nakazawa Ishihara Inooka 2003). K.K., eds. Medication safety: A guide for healthcare
3. A model for driver perception, which is an facilities. Bethesda, MD: American Society of Health
ambitious additional function of the system. The Systems Pharmacists; 2005: p. 73-88.
scope of this module is to maintain a Dingus, T.A., Antin, J.F., Hulse, M.C., and Wierwille,
representation of the items the driver is aware W.W. (1989). Attentional Demand Requirements of an
of, which do not necessarily coincide with the Automobile Moving-Map Navigation System,
Transportation Research , 23A (4), 301-315.
real world. For this module, the feasibility to the
Ehmanns, D., Spannheimer, H. (2004). Roadmap,
expected accuracy is not sure (see risks section),
available on www.adase2. Net
but it does not cost much (here) and if it works it European Communities (2009). EU energy and transport
may provide additional very useful information in figures, Statistical Pocketbook, Belgium
(e.g., understanding that a mistake comes form a It can be accessed through the Europa server
missing entity in driver world). The function (http://europa.eu).
combines eye gaze observations with Flemisch F., Adams C., Conway S., Goodrich K., Palmer
information from the perception layer to M., and Schutte P., (2003). The HMetaphor as
determine which objects and points the driver Guideline for Vehicle Automation and Interaction,
watched. The gazed features are introduced into (NASA/TM—2003-212672). Hampton, VA: NASA
an alternative representation of the world (the Langley Research Center.
driver mental model of the world) and then Green, P. (2000). “Crashes Induced by Driver
evolve according to rules that plausibly Information Systems and What Can be Done to Reduce
reproduce the assumptions a driver do about Them” (SAE Paper 2000-01-C008). In Convergence
objects he is no longer looking at. 2000 (pp. 26–36). SAE Publication
4. An interaction manager in the form of a Griffiths, P., Gillespie, R. (2004). Shared Control
variable plug-in. It will show how a variety of between Human and Machine: Haptic Display of
interactions, suited for different types of Automation during Manual Control of Vehicle
vehicles and different types of support, can all Heading, 12th International Symposium on Haptic
be built above the same unified situation Interfaces for Virtual Environment and Teleoperator
assessment produced by the core application. Systems (pp. 358-366). Chicago, IL, USA 2004.
Gulian, E., Matthews, G., Glendon, A. I., Davies, D. R.,
& Debney, L. M. (1989a). Dimensions of driver stress,
Ergonomics, 32, 585-602.
References Hill W., Stead L., Rosenstein M., and Furnas G. (1995).
Adomavicius G., Tuzhilin A. (2005). Toward the Next “Recommending and Evaluating Choices in a Virtual
Generation of Recommender Systems: A Survey of the Community of Use,” Proc. Conf. Human Factors in
State-of-the-Art and Possible Extensions, IEEE Computing Systems, 1995.
Transactions on Knowledge and Data Engineering 17 Hoch S., Schweigert M., Althoff F., Rigoll G. (2007).
(6) 2005, 734-749 “The BMW SURF Project: A Contribution to the
Andreone, L. Amditis A., Deregibus E., Damiani S., Research on Cognitive Vehicles,” in Proc. 2007 IEEE
Morreale D., and Bellotti F., (2005). “Beyond Context- Intelligent Vehicles Symposium, Istanbul, Turkey, June
Awareness: Driver-Vehicle-Environment Adaptively. 3-15, 2007.
From the COMUNICAR Project to the AIDE Ji, Q., Yang, X. (2002). Real-time eye, gaze, and face
Concept.” Proc. IFAC 16th World Congress, 2005 pose tracking for monitoring driver vigilance. Real-
Berghout, L., Versteegt, E., van Arem, B. (2003). Time Imaging (8), 1077–2014
Advanced driver assistance systems; Results of the Langley, P. (1999). "User Modeling in Adaptive
State of the Art of ADASE-II, available on Interfaces". In Proceedings of the Workshop on
www.adase2.net (consulted on 2sd September 2010). Machine Learning in User Modeling, Advanced
Bertollini GP, Hogan RM (1999). Applying Driving Course on Artificial Intelligence (ACAI ’99), Chania,
Simulation to Quantify Steering Effort Preference as a Greece, 1999
Function of Vehicle Speed. Society of Automobile (http://www.iit.demokritos.gr/skel/eetn/acai99/Worksh
Engineers, 1999-01-0394 ops.htm).
Cacciabue P.C. (1998). Modelling and Simulation of Lolli F., Marzani S., Minin L., Montanari R., Calefato C.
Human Behaviour in System Control; Springer-Verlag, (2009). “Not-invasive add-on functionalities for the
London, UK. realtime assessment of driver-vehicle interaction”, in
Catino M. (2002) Da Chernobyl a Linate. Incidenti Proceedings of VI Conference of the Italian Chapter of
tecnologici o errori organizzativi? Carocci editore, AIS, Achieving Fusion in the Interconnected World:
Roma Exploring the connection between organizations and
Console L., Torre I., Lombardi I., Gioria S. and Surano technology, Costa Smeralda (Italy), October 2-3, 2009
V. (2003). Personalized and adaptive services on Mattes, S. (2003). The lane change task as a tool for
board a car – an application for tourist information, driver distraction evaluation. In H. Strasser, H. Rausch
& H. Bubb (Eds.), Quality of work and products in
41
CEUR Proceedings 4th Workshop HCP Human Centered Processes, February 10-11, 2011
enterprises of the future (pp. 57-60). Stuttgart: Tango F., Botta M. (2009). Evaluation of Distraction in a
Ergonomia Verlag. Driver-Vehicle-Environment Framework: An
McDonald, W. A., & Hoffman, E. R. (1980). Review of Application of Different Data-Mining Techniques,
relationships between steering wheel reversal rate and Lecture Notes in Computer Science, 2009, Volume
driving task demand. Human Factors 22, 733-739. 5633, Advances in Data Mining. Applications and
McLean, J.R. & Hoffman, E.R. (1975). Steering wheel Theoretical Aspects, Pages 176-190
reversals as a measure of driver performance and Tango F., Calefato C., Minin L., Canovi L. (2009).
steering task difficulty. Human Factors 17,248-256. “Moving attention from the road: a new methodology
Minin L., Marzani S., Tesauri F., Montanari R., Calefato for the driver distraction evaluation using machine
C. (2009). "Context-dependent force-feedback steering learning approaches”, Proceedings of the 2nd
wheel to enhance drivers’ on-road performances”, in conference on Human System Interactions, May 21-23,
Proceedings of HCI International 2009, 9-24 July 09, 2009, Catania, Italy
San Diego, CA, USA Steele, M., & Gillespie, B. (2001) “Shared control
Nakazawa S., Ishihara T., Inooka H. (2003). Real-time between human and machine: Using a haptic steering
algorithms for estimating jerk signals from noisy wheel to aid in land vehicle guidance”, Proceedings of
acceleration data, Int.l J. of Applied Electromagnetics the Human Factors and Ergonomics Society 45th
and Mechanics, 18, 1-3, pp. 149-163, 2003. annual meeting (pp. 1671– 1675). Minneapolis,
Neale V. L., Dingus T. A., Klauer S. G., Suweeks J., Minnesota, USA 2001.
Goodman M. (2005). An overview of the 100-car Tango F., Minin L., Tesauri F., Montanari R (2010).
naturalistic study and findings, Paper Number 05- Field tests and machine learning approaches for
0400, National Highway Traffic Safety refining algorithms and, correlations of driver’s model
Administrationin,Washington D.C. parameters, Applied Ergonomics Volume 41, Issue 2,
Neisser U. (1967). Cognitive Psychology. Appleton- March 2010, Pages 211-224
Century-Crofts, New York Verwey, W.B. (1991). Towards Guidelines for in-car
Perry S.J., Wears R.L., Cook R.I. (2005). The role of Information Management: Driver Workload in Specific
automation in complex system failures. J Patient Safety Situations. Technical Report IZF 1991 C-13,
2005; 1: 56-61. Soesterberg, The Netherlands: TNO Institute of
Regan M. A, Lee J. D., Young K. L. (2009). Driver Perception.
Distraction. Theory, Effects, and Mitigation, CRC Young, K. & Regan, M. (2007). “Driver distraction: A
Press, Taylor & Francis Group, New York. review of the literature”. In: Faulks I.J., Regan M.,
Resnick P., Iakovou N., Sushak M., Bergstrom P., and Stevenson M., Brown J., Porter A. & Irwin J.D. (Eds.).
Riedl J. (1994). “GroupLens: An Open Architecture for Distracted driving. Sydney, NSW: Australasian
Collaborative Filtering of Netnews,” Proc. 1994 College of Road Safety.
Computer Supported Cooperative Work Conf., 1994.
Shardanand U. and Maes P. (1995). “Social Information
Filtering: Algorithms for Automating ‘Word of
Mouth’,” Proc. Conf. Human Factors in Computing
Systems, 1995.
Tan, P.-N. (2005). Introduction to Data Mining. Pearson
Addison Wesley, Boston.
42