You are on page 1of 14

Cognition, Technology & Work (2019) 21:657–670

https://doi.org/10.1007/s10111-019-00559-2

ORIGINAL ARTICLE

Driver–vehicle cooperation: a hierarchical cooperative control


architecture for automated driving systems
Chunshi Guo1,2   · Chouki Sentouh1 · Jean‑Baptiste Haué2 · Jean‑Christophe Popieul1

Received: 31 July 2017 / Accepted: 6 April 2019 / Published online: 27 April 2019
© Springer-Verlag London Ltd., part of Springer Nature 2019

Abstract
The concept of automated driving changes the way humans interact with their cars. However, how humans should interact
with automated driving systems remains an open question. Cooperation between a driver and an automated driving sys-
tem—they exert control jointly to facilitate a common driving task for each other—is expected to be a promising interaction
paradigm that can address human factors issues caused by driving automation. Nevertheless, the complex nature of auto-
mated driving functions makes it very challenging to apply the state-of-the-art frameworks of driver–vehicle cooperation
to automated driving systems. To meet this challenge, we propose a hierarchical cooperative control architecture which is
derived from the existing architectures of automated driving systems. Throughout this architecture, we discuss how to adapt
system functions to realize different forms of cooperation in the framework of driver–vehicle cooperation. We also provide
a case study to illustrate the use of this architecture in the design of a cooperative control system for automated driving. By
examining the concepts behind this architecture, we highlight that the correspondence between several concepts of planning
and control originated from the fields of robotics and automation and the ergonomic frameworks of human cognition and
control offers a new opportunity for designing driver–vehicle cooperation.

Keywords  Driver–vehicle cooperation · Human–machine cooperation · Shared control · Hierarchical architecture ·


Automated driving

1 Introduction the role of the human driver. What should be the appropri-
ate interaction paradigms between the driver and the system
Automation systems have been around automobiles for a constitutes a key question, since a large body of research has
long time. For example, the first cruise control system was pointed out that potential problems with the automation are
introduced by Chrysler in the 1950s (Weltmore 2003). often related to human factors (Bainbridge 1983; Endsley
Research on automated driving technologies has also been and Kiris 1995).
carried out from way back, including the European project The taxonomy of SAE international (SAE 2016) speci-
PROMETHEUS (Dickmanns 1986) and the American pro- fies the allocation of driving tasks between the driver and
ject PATH (Shladover et al. 1991) during the 1980s and the system at each level of automation. In the scope of the
1990s. But it was not until recently that the demonstra- taxonomy of SAE, many studies on human–machine inter-
tions of DARPA Urban Challenge (Buehler et al. 2009) and action of automated vehicles have focused on the transition
Google Car (Markoff 2010) have spurred significant inter- from automated to manual control (Blanco et al. 2015). Par-
est in automated driving. Driving automation engenders a ticularly, the transition initiated by a level 3 system when
shift of functions to the system that significantly changes it reaches the limit of its operating conditions has received
a great degree of attention (Gold et al. 2013; Lorenz et al.
* Chunshi Guo 2014; Walch et al. 2015). Nevertheless, the taxonomy of
chunshi.c.guo@renault.com SAE which is elaborated from an engineering point of view
oversimplifies the interaction between the driver and the sys-
1
LAMIH-CNRS UMR 8201, University of Valenciennes, tem by considering it as an “all-or-nothing” function alloca-
Mont Houy, 59313 Valenciennes, France
tion, i.e., the system performs the entire driving task when it
2
Cognitive Ergonomics and HMI, Technocentre Renault, 1 is engaged and the driver drives manually when he/she takes
Avenue du Golf, 78280 Guyancourt, France

13
Vol.:(0123456789)

658 Cognition, Technology & Work (2019) 21:657–670

over the control. In the literature, few works proposed new These considerations motivate us to apply driver–vehi-
interaction paradigms for automated driving systems1 other cle cooperation to the design of automated driving systems.
than the “all-or-nothing” function allocation presumed by Theoretical models and frameworks from the domain of
the taxonomy of SAE. human-centered automation can be employed to guide the
The idea of forming a cooperation teamwork between the design of driver–vehicle cooperation. For instance, shared
human operator and the automation has been proposed since control has been successfully applied to driver assistance
the introduction of automated systems in various domains systems. How to apply these general frameworks which
(Hoc et al. 1994). In the context of driving automation, usually take an ergonomic viewpoint to an automated driv-
driver–vehicle cooperation (also referred to as cooperative ing system is not trivial, because the latter is composed of
guidance and control by Flemisch et al. (2014)) is gaining sophisticated functions covering from perception, planning
recognition as a promising interaction paradigm for in-car (or decision making) to control, in contrast with a driver
automated systems. According to Hoc, a human operator and assistance system, e.g., lane-keeping assist, where coopera-
an automated system can attain cooperation if they meet two tion usually occurs at the control level. Therefore, a func-
minimal conditions: tional architecture of automated driving system is necessary
to modularize complex system functions. However, most
1. Each one strives towards their own goals and can inter- architectures found in the literature have been designed for
fere with the other one’s goals, resources, procedures, full automated driving and do not support a human–machine
etc. interaction on a cooperative basis. A main goal of this paper
2. Each one tries to manage the interference to facilitate is to propose a hierarchical cooperative control architecture
the individual activities and/or the common task when for an automated driving system that can cooperate with a
it exists (Hoc 2001). human driver. Through this architecture, we intend to inves-
tigate how to adapt planning and control approaches through
Studies have demonstrated that driver–vehicle coopera- shared control (Abbink et al. 2012) and human supervisory
tion is beneficial for the driver’s performance in circum- control (Sheridan 1992) to achieve cooperative driving.
stances such as situation awareness and control performance Another research question concerns how to apply principles
during the interaction with the system. We argue that the and guidelines for driver–vehicle cooperation proposed from
driver–vehicle cooperation can also contribute to user expe- the community of human-centered automation to system’s
rience and system performance in the case of automated planning and control function design which is rather techni-
driving. According to a recent online survey on the accept- cally orientated.
ance of autonomous vehicles with 1000 German respond- To reach these goals, Sect. 2 reviews several shared con-
ents, “the possibility of reassuming control of the vehicle trol frameworks, human supervisory control implementa-
or terminating the automated driving procedure at any time tions and ergonomic frameworks for driver–vehicle coop-
is one of the central needs” of the respondents and “only eration. By discussing the benefits and limitations of these
a minority would wish to cease paying attention to traffic frameworks, we point out the need for a general functional
and completely cede control of the vehicle to the automated architecture that describes how to involve automated driving
system” (Wolf 2016). Driver–vehicle cooperation enables functions in a desired cooperation framework. In Sect. 3,
the system to share vehicle control. Instead of disengaging we propose a hierarchical cooperative control architecture
the system, the driver can directly modify vehicle’s behav- to accommodate this need. At each level, we propose new
ior exerting control at the same time. On the human side, approaches to achieve the proposed cooperation form. Sec-
driver–vehicle cooperation could satisfy potential user needs tion 4 presents a case study in which we instantiate this
and enhance the user experience. On the system side, the architecture in a use case concerning highway merging
system could benefit from the help of the driver to better management. Finally, we conclude this paper by summariz-
handle interactions with other road users, particularly con- ing the contributions of this architecture and discussing its
sidering that these interactions sometimes exhibit strong limitations in Sect. 5.
social patterns, e.g., cues like eye contacts and hand ges-
tures, which are difficult for a machine to interpret (Färber
2016). 2 Prior work on driver–vehicle cooperation

2.1 Shared control
1
  According to SAE’s taxonomy, automated driving systems refer to
A shared control framework first emphasizes that both the
those systems that are capable of performing all driving tasks without
human’s monitoring, covering the highest three levels of automation human operator and the system perform control simulta-
(from Level 3 to Level 5). neously, thus making it possible to realize a seamless and

13
Cognition, Technology & Work (2019) 21:657–670 659

smooth shift of authority. It represents an important advan- increased its control when the driver was in underload and
tage compared with the paradigm of “switch on/off” result- overload conditions.
ing in a binary shift. Another feature of shared control is Although haptic shared control has been proved to enhance
that the human and the system can communicate their goals driver’s control performance, it suffers from potential conflicts
directly on the control interface. As pointed out by some between the driver and the system, which could increase driv-
theoretical works (Flemisch et al. 2010, 2016), the concept er’s physical efforts and even result in accidents (de Winter
of shared control is coherent with driver–vehicle coopera- and Dodou 2011). Haptic shared control methods proposed in
tion (from this perspective, shared control can be referred to the literature today mitigate the conflicts in a reactive manner,
as cooperative control) considering that the driver and the i.e., by reducing the controller’s control input when conflicts
system share their control to reach a common goal. arise or by applying a kind of arbitration, i.e., deciding whose
Depending on the type of control interface, shared control control input finally acts on the controlled object (Kelsch
approaches can be classified into two categories (Abbink et al. 2006; Baltzer and Flemisch 2017). However, this limit
and Mulder 2010): blended shared control and haptic shared of haptic shared control is due to the controller that relies on a
control. Blended shared control is implemented on a drive- specific path and needs a specific goal. When the driver has a
by-wire interface allowing a mechanical decoupling of the different goal, the two enter into a conflict situation. As shown
control interface and the controlled plant. Therefore, the later in this paper, high-level functions in an automated driving
driver’s control, in form of electronic signal, can be blended system can be used to address this problem.
with that of the controller flexibly. Andersen et al. (2013)
used a weighting function of the front tire slip angle to blend 2.2 Human supervisory control
the control inputs of a driver and an automatic steering con-
troller such that more authority is allocated to the control- In a human supervisory control framework, the human
ler when the vehicle begins to lose stability. Sentouh et al. operator provides specific goals or waypoints for the system
(2013) developed a shared steering assist controller which (Sheridan 1992). In contrast with shared control scheme
blends driver’s control based on a metric characterizing the where human input is continuous, a human supervisor inter-
conflicts between the driver and the system. Erlien et al. mittently intervenes in the control process to give a new
(2016) formulated a blended shared control problem in the goal. Widely applied for telerobotic and process control,
framework of model predictive control (MPC). The control- human supervisory control is gaining interest in the auto-
ler seeks to identically match the human’s control in normal motive community following the progress of driving auto-
conditions and adapts its control to prevent hazards only if mation. A representative example is the so-called “Active
the predicted vehicle trajectory goes out of a safety zone. Lane Change Assist” application which has already become
In a blended shared control framework, the driver may commercially available (Quain 2016). In this application,
not be aware of the system’s activity, because the blending the driver can initiate a lane change maneuver performed
occurs in the system controller. To overcome this limit, a by the system by switching the turn signal. The project of
haptic shared control scheme emphasizes that the driver and Conduct-by-Wire (Geyer 2013) proposed and implemented a
the system exert forces on the control interface such that the new concept that the driver assigns maneuver commands via
driver can continuously feel the activity of the system (Abbink a so-called maneuver interface that enumerates all possible
et al. 2012). The control force of the controller can be used maneuvers in the applied use cases, and then the system
as a guidance force (Mulder et al. 2011) or as a channel to performs the assigned maneuvers automatically.
convey the environmental hazards to the drive (Takada et al. Human supervisory control scheme offers convenience
2013; Guo et al. 2015). With these design perspectives, the for the driver by allowing him/her to express directly his/her
system’s control force is usually computed in the framework intention to the system through symbolic command. How-
of impedance control (Hogan 1985), to achieve the desired ever, a main issue in human supervisory control is human’s
characteristics of impedance, e.g., a target stiffness of the con- out-of-the-loop performance problem (Endsley and Kiris
trol interface. There are also studies focusing on the author- 1995). What’s more, drivers may still have the need to subtly
ity allocation. Saleh et al. (2013) and Soualmi et al. (2014) control vehicle’s trajectory by turning the steering wheel or
formulated a predefined compromise between lane-keeping pressing pedals, which cannot be met by human supervisory
quality and control sharing as an optimization objective of an control.
optimal automatic steering controller. Petermeijer et al. (2015)
performed experiments to compare continuous versus band- 2.3 Frameworks for driver–vehicle cooperation
width haptic steering guidance on driver’s lane-keeping perfor- from a cognitive perspective
mance. Nguyen, Sentouh, and Popieul (2016) modulated the
torque of an automatic steering controller based on the driver’s If shared control and human supervisory control can be
involvement level and drowsiness state. Thus, the controller viewed as frameworks for human–machine cooperation

13

660 Cognition, Technology & Work (2019) 21:657–670

Fig. 1  A typical functional Percepon Planning & control


& situaon assessment
architecture of AD systems
(solid-border blocks for func-
tions, while dash-border blocks • Digital map Route
• Traffic informaon planning
for interfaces)

Replan Route
request segment

• Traffic rules Strategical level


SENSOR • Context informaon Maneuver
LAYER • Spaal awareness planning Taccal level
• Scene evoluon predicon
Operaonal level

Replan
Maneuver plan
request
Reference Actuator inputs
• Road geometries trajectory Vehicle
Moon ACTUATOR
• Object states planning control LAYER
• Safe travel field

that stem from the application of conventional control 3 Proposition of a hierarchical cooperative
theory, then such frameworks put emphasis on the cogni- control architecture
tive aspects (perceiving, internal representation, decision
making) of the human operator and aim to cover the gen- An automated driving system consists of three main func-
erality of human–machine cooperation. tionalities—perception, planning and control (Taş et al.
The framework proposed by Hoc (2001) views 2016). To interact asynchronously in real time with an
human–machine cooperation as an activity of interfer- uncertain and dynamic driving environment, automated
ence management. It also emphasizes the role of a shared driving systems commonly adopt layered system archi-
representation of the task environment maintained by the tecture in which each layer operates at a characteristic
actors in cooperation. Flemisch et al. (2014) proposed a frequency with a certain level of task abstraction (Korten-
“cluster concept” englobing a list of attributes that char- kamp and Simmons 2008). An interesting finding is that
acterize the quality of cooperation. These two frameworks most of the layered architecture of automated vehicle pro-
provide the following important principles in driver–vehi- totypes shares the same hierarchy with the hierarchical
cle cooperation design: dynamic task distribution, inter- model of Michon. Examples are the PRORETA 3 project
ference management (arbitration of conflicts) and main- (Bauer et al. 2012) and PROUD project demonstrators
taining a shared representation (inner compatibility). (Broggi et al. 2015) and Audi’s A7 concept car (Ulbrich
Beyond the above-mentioned frameworks that intend and Maurer 2015). Figure 1 illustrates a typical layered
to define what “cooperation” is, some works decom- architecture for automated driving systems in which func-
pose cooperation into hierarchical levels to reduce the tions are mapped into the tri-level hierarchy of Michon.
complexity of the cooperation design (Hoc et al. 2009; The hierarchical models of driver–vehicle cooperation
Flemisch et  al. 2014; Pacaux-Lemoine and Flemisch (cf. Sect. 2.3) reveals the possibility of decomposing coop-
2016). These hierarchical models commonly inherit the eration. We also notice that the planning and control func-
hierarchy of the model proposed by Michon (1985) which tions used for automated driving share similar principles
decompose the driving task into strategical, tactical and with models for driver–vehicle cooperation in the litera-
operational levels. Recently, efforts are underway to unify ture. This aspect makes it possible to extend the planning
these tri-level hierarchical models and to propose consist- and control bloc at each level of the system architecture
ent terms (Flemisch et al. 2016). (Fig. 1) to achieve a specific cooperation form. In this
These ergonomic frameworks provide a variety of way, a new architecture for planning and control function,
qualitative characterizations of driver–vehicle coopera- referred to as hierarchical cooperative control architec-
tion. However, they do not provide a technical solution ture, is derived, as shown in Fig. 2. The system’s planning
to implement these frameworks into the development function at either the strategical or tactical level is usually
of functions of automated driving system. A functional designed to arbitrate a plan (route segment or maneuver)
architecture that bridges the gap between these conceptual from multiple alternatives, i.e., only one plan selected at
frameworks and the technical development of a coopera- a time will be executed by the control level for a period
tive system is still needed.

13
Cognition, Technology & Work (2019) 21:657–670 661

Percepon Hierarchical cooperave control HMI


& situaon assessment

Route
• Digital map Route alternaves Route Route alternaves Navigaon
• Traffic informa on planning arbitraon interface
Arbitraon Desnaon /
result Selected route
Replan Route fragment
Maneuver plan
request

Maneuver Maneuver Mulmodal


• Traffic rules
alternaves alternaves feedback
SENSOR • Context informa on Maneuver Maneuver Maneuver
LAYER • Spa al awareness planning arbitraon interface
• Scene evolu on predic on Arbitraon Driver’s plan Driver
result commands DRIVER
Replan Maneuver plan Maneuver
request intenon
Reference
• Road geometries trajectory Hapcs Vehicle
Moon Shared
• Object states control
planning control
• Safe travel field Predicted Driver’s control interface
trajectory

Actuator inputs
Strategical level
ACTUATOR
LAYER
Tac cal level
Opera onal level

Fig. 2  Architecture of the hierarchical cooperative control within an AD system (solid-border blocks for functions, while dash-border blocks for
interfaces)

of time until another plan is selected (Pirjanian 1999). A economy and available time for using the automated driv-
similar principle of “arbitration” has also been proposed ing mode. Whereas the arbitration on the route alternatives
for conflict management in the scope of driver–vehicle can simply follow driver’s selection (or his/her recorded
cooperation (Kelsch et al. 2006). Hence, we propose an preference) before the trip, but the arbitration should take
arbitration form for the cooperation at the strategical and into account more factors if the driver intervenes in the
tactical levels. Given the continuous nature of the control vehicle control loop during the trip. The driver can directly
input (whether it comes from the system or the driver) change into a new route or a new destination via the naviga-
and the success of shared control in managing the shift of tion interface that is shown in Fig. 2. The route arbitration
control authority, the cooperation issue in the operational can also be influenced by cooperation at lower levels. For
level of this architecture is formulated as a shared control instance, when the driver intervenes at the operational or
problem. tactical level to make the vehicle move towards a direction
The driver’s commands or control enter the architecture not planned by the system at a highway split, the system
via a proper HMI at each level. Without the driver’s inter- needs to re-plan a route or ask the driver whether he/she
vention, the system’s own decision and control at each level wishes to change to another destination.
are passed directly to its subordinate level, i.e., the system Strategical decisions on route usually take place in min-
operates in the same way as in a classical architecture as utes during the trip. Moreover, such decisions are much less
shown in Fig. 1. This architecture offers a generic descrip- relevant to the immediate safety compared with those at the
tion of system modules and inter-module communications other two levels. Perhaps for these reasons, cooperation at
involved in driver–vehicle cooperation. In the following sec- the strategical level has been scarcely investigated in the
tions, we propose new approaches to achieve the cooperation framework of driver–vehicle cooperation.
form at each level.
3.2 Cooperation at the tactical level
3.1 Cooperation at the strategical level
The tactical level generates maneuver plans not only to
At the strategical layer, the driver and the system cooperate follow the planned route but also to handle local driving
to plan a route to a destination. Similar with nowadays in-car situations, e.g., to overtake a slow vehicle or to yield at an
navigation systems, route planning searches multiple route interaction. This task on the system side is referred to as
candidates based on criteria such as time saving, energy maneuver planning. It is interesting to view that several

13

662 Cognition, Technology & Work (2019) 21:657–670

maneuver planning techniques applied to automated driv- latter can select an alternative if he/she is not satisfied with
ing can be explained by a cognitive model proposed by the system’s intention.
Johannsen and Rouse (1979) for human operator’s plan- The second one to realize maneuver cooperation exploits
ning activities. In this model, a rational planning process the inter-level communication of the architecture. The sys-
was broken down into four steps: (1) generating feasible tem infers the driver’s intention when he/she acts on the
alternatives, (2) imaging the consequences, (3) valuing control interface (steering wheel or pedals) on the opera-
(or evaluating) the consequences of the alternatives, (4) tional level. In this sense, this method integrates shared
choosing an alternative as the plan. To generate feasible control in maneuver cooperation. Different techniques for
alternatives in step (1), there exist principally two groups the inference of the driver’s maneuver intention exist in
of approach in the field of robotics. The first group fol- the literature based on cues from vehicle dynamic states,
lows the behavior-based paradigm that highlights which driver’s behavior, road configuration and traffic information
maneuver to perform strongly depending on the current (Kumagai 2006). Once the driver’s intention is inferred, the
situation (Matarić and Michaud 2008). Therefore, basic system can adapt the maneuver plan to that inferred one in
maneuvers are usually designed a priori considering the maneuver arbitration. This cooperation method ascends the
situations to be addressed and the goal to be attained by interference at the operational level to the tactical level.
each behavior. For example, finite state machine, a behav-
ior-based formalism that was overwhelmingly used by the 3.3 Cooperation at the operational level
autonomous vehicle porotypes in the DARPA Challenge to
model maneuver planning strategy (Buehler et al. 2009). The driving task at the operational level is to conduct vehi-
The second group relies on sampling-based planning tech- cle steering and speed control to execute the maneuver plan
niques which were originally applied to motion planning generated from the tactical level. On the system side, this
and obstacle avoidance. Approaches of this group decom- task is usually performed by motion planning and trajectory
pose a feasible trajectory space into discrete subspaces (or control. Motion planning first plans a trajectory or a path as a
homotopies) in real time (Bender et al. 2015; Park et al. realization of a maneuver plan (usually as end vehicle poses
2015; Gu et al. 2016). Each subspace can be viewed as a or states). With the driver’s intervention, a shared control
maneuver alternative. Step (2) corresponds to predicting is employed to determine the final control input on vehicle
the trajectories of the detected dynamic objects. In step actuators.
(3), the values of the consequences of the alternatives can In addition to the classical shared control techniques
be modeled by costs, risks or utilities using correspond- presented in Sect. 2.1, the introduction of motion planning
ing mathematical tools. Step (4) corresponds to maneuver can inspire new approaches to implement shared control.
arbitration, i.e., to select that alternative with the conse- Assuming that the conflict between the driver and the sys-
quences with the greatest positive value. tem is mainly due to the mismatch of the system’s reference
Since a driver may have a different anticipation (step 2) trajectory with that of the driver, the first approach aims at
of the driving situation or different value criteria (step 3) developing a motion planner that adapts the planed trajec-
than that of the system, he/she may interfere with the system tory to driver’s control. This approach was explored in the
on the maneuver plan (Löper and Flemisch 2009). In this work of Brandt et al. (2008) where a path planner places a
hierarchical architecture, two means of interaction can be series of elastic band nodes to an extrapolated vehicle path
exploited to manage this interference. based on the driver’s current steering angle. In the recent
The first one is offering the driver a means to directly French project CoCoVeA, Benloucif et al. (2017) developed
indicate his/her maneuver plan at the tactical level, which a model that predicts the driver’s desired lateral displace-
follows a human supervisory control scheme. This kind of ment within a lane for a motion planner.
maneuver-based interaction has been explored, e.g., in the Over the last decade, MPC has been applied to motion
project HAVEit where the system recommended maneuvers planning and vehicle control in the application of automated
to the driver through a “maneuver–grid” interface (Glaser driving. Two features of MPC make it appealing as a frame-
et al. 2010) or in the concept of “Conduct by Wire” in which work to accommodate shared control. First, it is possible
the driver was supposed to be a supervisor to routinely gen- to unify motion planning and trajectory control to a single
erate or update a plan for the system (cf. Sect. 2.2). However, constrained finite horizon optimal control problem. In this
a recent study which compared four maneuver interaction way, a MPC controller can compute commands toward a
concepts suggested user’s preference to transfer as many desired terminal state (maneuver plan) while respecting the
as tasks as possible to the system (Albert et al. 2015). To constraints. In this sense, the strategy of MPC is closed to
relieve the driver from the supervising task, a possible inter- human driving behavior assumed by Gibson and Crooks
action way could be that the automation shows its intended (1938), that humans tend to control the vehicle within a field
maneuver and plausible alternatives to the driver, while the of safe travel rather than obsessively track a single path.

13
Cognition, Technology & Work (2019) 21:657–670 663

Fig. 3  A typical driving scene


on highway entry section
・Ego vehicle ・Lead vehicle

AD
・Merging vehicle

Second, MPC iteratively resolves an optimization problem damping, is the steering ratio, Tal the total torque of auto-
online. Thus, integrating the driver’s actual control into the alignment of two front wheels and kp the steering assist coef-
optimization problem, MPC can achieve dynamic allocation ficient which is a simplified representation of the electric
of control authority. motor’s assistant steering torque (Mitschke and Wallentow-
Here we present a novel MPC framework to accommo- itz 2004). Equation 3 means that the driver’s torque ( Td )
date shared control for vehicle trajectory tracking. Consider and that of the system ( Tctrl ) are superposed on the steering
such a constrained optimal control problem: wheel.
The sets , f and  denote the constraints on states,
Minimize:JNp (𝐱(0), U(0)) = 𝜌Jout + Jin (1a) terminal states and input, respectively. These constraints
Subject to: can also be exploited in a shared control scheme. The state
constraints can be set on the vehicle’s position to define a
̇ + 1) = f (𝐱(k), 𝐮(k)),
𝐱(k k = 0, … , Np − 1, (1b)
field of safe travel such that the MPC will enforce its control
𝐱(k) ∈ , k = 0, … , Np − 1, (1c) to constrain the vehicle’s trajectory. If the control input is
steering torque, it can also be used as a channel to convey
( ) the potential hazard to the driver. If the final authority is
𝐱 Np ∈ Xf , (1d)
accorded to the driver, the control constraint  ensures that
the driver can always override the system’s control (steering
𝐮(k) ∈  , k = 0, … , Nc − 1. (1e) torque). Together with the state constraints, such a control
The cost function JNp (1a) incorporates two competing constraint can implement the so-called “soft protection”
objectives—small output deviation Jout and small control concept proposed by Itoh and Inagaki (2014). Otherwise,
efforts Jin . The parameter 𝜌 gives the relative weighting the control constraint can be set high enough that the driver
between Jout and Jin , and can be used to leverage the alloca- cannot overrule the maximal system’s control to realize the
tion of the control authority between the system and the “hard protection” concept.
driver. The smaller 𝜌 is, the more the MPC controller can MPC can also be adapted to implement blended shared
tolerate the deviation from the reference and the more it can control. One example is the work of Erlien et al. (2016)
penalize using control authority. Consequently, more control which is briefly presented in Sect. 2.1.
authority is left to the human driver to adjust the trajectory.
Inversely, when 𝜌 becomes bigger, the controller will employ
more control authority to penalize the deviation from its 4 Case study: highway merging
reference. (1b) is the plant model used for prediction with management
Np as the prediction horizon. To formulate a haptic shared
control problem, we choose steering torque as the control 4.1 Use case and concept for cooperation
input 𝐮(k) . The control interface is hence the steering system
whose dynamics can be modeled as In this section, we explain how this general hierarchical
Tal cooperative control architecture can be used to drive the
Ieq 𝛿̈f + ceq 𝛿̇ f = is 𝐮(k) − (2) design of a cooperative control system (as a sub-system of an
kp
automated driving system) through a case study. We define
a use case in which the driver and the system cooperate to
𝐮(k) = Td + Tctrl , (3) handle a merging situation at a highway entry. As shown
where Ieq is the equivalent moment of inertia of the steer- by Fig. 3, the ego vehicle (automated vehicle) encounters a
ing system, 𝛿f the front steering angle, ceq the equivalent merging vehicle at a highway entry section. Studies on traffic

13

664 Cognition, Technology & Work (2019) 21:657–670

Situaon assessment Hierarchical cooperave control HMI


Maneuver
Mulmodal
Maneuver alternaves Maneuver feedback
alternaves cooperaon
• Context informaon Maneuver
Transion Driver’s selecon interface Driver’s
• Spaal awareness state
• Scene evoluon predicon Arbitraon supervisor selecon
machine result
Driver’s
SENSOR intenon
LAYER detecon Driver’s
Replan Maneuver Detected steering torque
request plan intenon
DRIVER
• Road geometries Steering Steering torque
• object states
MPC-based shared steering control
System’s wheel feedback
steering torque
Steering torque of
angle auto-alignment
ACTUATOR
Driver’s control System’s control Cooperaon result LAYER

Fig. 4  Functional architecture of the cooperative control system proposed for the use case of highway merging management

management (Hidas 2005) showed that merging–yielding process are highlighted. Note that the driver and the system
interaction between vehicles on merging section exhibits cooperate on vehicle steering control at the operational level
social patterns. As usually observed in daily life, a mainline in the current design.2 Therefore, automatic vehicle longitu-
vehicle may change to its left lane to yield to a merging dinal control functions are not shown in Fig. 4 for clarity. We
vehicle with courtesy. Sometimes, the driver in a merging present the development of each function briefly as follows.
vehicle may invite a mainline vehicle to pass by deceler-
ating a little or by eye contact. If it appears difficult for a 4.2.1 Situation assessment
“machine” to interpret those social cues, the driver could
help the system handle this kind of social interaction via A highway merging situation is contextual and dynamic,
driver–vehicle cooperation. thus it is crucial for the system to understand and assess the
In this use case, we designed two ways allowing the driver current situation. As shown in Fig. 4, the situation assess-
to cooperate with the system to handle a merging vehicle. ment function generates two levels of abstraction of the
The first way to cooperation which was proposed in our pre- external environment. At the tactical level, a high precision
vious work (Guo et al. 2017a) occurs at the tactical level. digital map is first used to give contextual information. Next,
When the system detects a vehicle whose merging may influ- the system maintains a matrix-form spatial representation in
ence ego vehicle’s current state, it initiates cooperation by which the surrounding traffic vehicle is mapped into matrix
showing its intended maneuver and plausible alternatives to cases with semantic meanings, e.g., “left leader” (Xiong
the driver. The driver can select an alternative if he/she does et al. 2016). In this way, the system can easily identify target
not agree with the intention of the system. In the absence of vehicles for maneuver planning. Lastly, once the target vehi-
the driver’s indication, the system will execute its intended cles are identified, the trajectories of these vehicles are pre-
maneuver if the current situation allows it. In the second dicted in a long-term prediction horizon (Guo et al. 2016).
way, cooperation arises from the operational level. If the The predicted trajectories are used to assess the risk of each
driver anticipates a merging situation earlier than the system, maneuver plan. At the operational level, the states of traffic
he/she can directly turn the steering wheel to initiate a coop- vehicles and the ego vehicle are represented in the Frenet
erative lane change without waiting the system to indicate frame (Willmore 2012) to decouple vehicle’s longitudinal
its intention. If a lane change is not feasible, e.g., due to the motion and lateral motion.
presence of a traffic vehicle in the target lane, the system From the perspective of driver–vehicle cooperation, the
renders resistance on the steering wheel to warn the driver. contextual information generated by situation assessment
will be fed to the HMI to maintain a shared representation
4.2 System architecture and function design with the driver.

As this use case concerns a local driving situation, the strate-


gical level which is relevant to global planning is not treated
here for simplicity. Figure 4 shows the functional architec- 2
  The driver and the system can of course cooperate in vehicle lon-
ture of the cooperative control system derived from the gen- gitudinal control in this use case as presented in our previous work
eral architecture. The information flows in the cooperation (Guo et al. 2015).

13
Cognition, Technology & Work (2019) 21:657–670 665

(a) (b)

AD

AD AD AD

Fig. 5  Lane-based constraints on vehicle’s lateral offset: a the constraints are placed on the outermost lane borders to prevent road departure; b if
the merge in an adjacent lane causes potential dangers, the border to this lane is set as constraint

4.2.2 Cooperation at the tactical level is responsible for detecting driver’s lane change intention
from his/her steering actions.
The first way of cooperation involves maneuver state The MPC-based shared steering control framework
machine and transition supervisor, two standard techniques originally presented in (Guo et al. 2017b) formulates the
to realize maneuver planning and maneuver arbitration in steering control sharing as the constrained optimal control
the general architecture. problem as presented in Sect. 3.3. The control authority
Maneuver state machine is a hierarchical finite state allocation is governed by the weighting parameter 𝜌 (1a)
machine (HFSM) that provides a simple and structured and the state constraints (1c) of the MPC framework. The
approach to model system’s tactics in this use case. The first- variation of 𝜌 is formulated as
level states represent two phases in the interaction with a {
𝜌 = 0, if 𝛾sw > 0 ∧ ||Tdr || > Tthre
merging vehicle. In the “intention phase”, the system adjusts . (4)
𝜌 = 1, if 𝛾sw = 0 ∨ 𝛾ma = 1
the ego vehicle’s relative positions to the merging vehicle
to manifest its intention, while the “decision phase” sys- The first “if” condition signifies the detection of the
tem engages an action to terminate this interaction. At the driver’s presence in the control loop. The variable 𝛾sw is the
second level, each state represents a maneuver plan—a tar- output of a steering wheel hand position sensing system
get state for motion planning and control at the operational with a positive value for the detection of at least one hand
level. There are three maneuver plans in “highway merging on the steering wheel. Tdr is the measure of the driver’s
management”—“pass”, “yield” and “lane change”. steering torque and Tthre is a threshold constant accounting
Transitions among system states are modeled in transition for the situation where the driver puts his/her hands on the
supervisor. While the system is designed to adopt conserva- steering wheel while loosely following wheel’s rotation.
tive tactics, i.e., transition supervisor triggers the transition The second “if” condition determines whether the sys-
to a “maneuver plan” with the lowest risk in the “intention tem exerts full control. There are two cases. The first one
phase”, the driver can trigger the transitions to other maneu- is when the driver releases the steering wheel ( 𝛾sw = 0 ).
ver plans through the maneuver cooperation interface. In this The second case corresponds to the cooperation method
way, potential interference between the driver and the system through which the driver initiates a lane change maneuver
on maneuver plan can be resolved. Note that this cooperation by turning the steering wheel. Driver’s intention detection
method is only available in the “intention phase”, since the at the tactical level places a detector on the driver’s steer-
system transitions to the “decision phase” when the situation ing torque and vehicle’s lateral and heading deviations to
is clear and the change of maneuver plan is either infeasible the lane center. Once the detection of driver’s lane change
or could hinder the immediate safety. intention is declared and this lane change is deemed as
The driver can also trigger transitions among maneuver feasible by situation assessment, the flag 𝛾ma is set to one
plans via cooperation at the operational level as presented which enables the system to exert control. Meanwhile, at
in the next section. the tactical level, the transition supervisor will trigger a
corresponding lane change maneuver for the MPC.
The ego vehicle’s position needs to be constrained in a
4.2.3 Cooperation at the operation level safe navigational zone for collision avoidance. In this use
case, we use lane-based constraints as a simplified form.
At the operational level, motion planning and shared con- The outermost lane borders are used as default lateral con-
trol functions of the general architecture are combined into straints. As illustrated in Fig. 5a, this placement allows the
a single MPC-based shared steering control framework. driver to perform a change lane. However, if an adjacent
Moreover, driver’s intention detection at the tactical level

13

666 Cognition, Technology & Work (2019) 21:657–670

Highway merging context Simulated augmented reality

The merging vehicle is highlighted


The HMI indicates:
• Merging scene representaon Buon command interface
• AD intended maneuver: yield
(blue triangles) The green-twinkling buon
indicates the availability of
• Available alternaves: pass & the alternaves
le lane-change (green triangle)

Fig. 6  HMI design for driver–vehicle cooperation in the use case of highway merging management

lane is occupied by vehicles and merging in this lane may • Button command interface: four capacitive backlit but-
result in collisions with in-lane vehicles, the constraint is tons allow the driver to express his/her intended maneu-
temporally set on the border of the occupied adjacent lane, ver to the system. These buttons are organized in four
as shown in Fig. 5b. The decision for constraint placement directions with up/down buttons for “pass/yield” and
is made during the assessment of the lane change feasibil- left/right for “left/yield lane change”. If a maneuver is
ity by situation assessment. If the driver intends to steer available, the corresponding button twinkles in green and
the vehicle out of the constraints on lane border, the MPC remains active. Pressing on an active button will trigger
enforces its control to constrain the vehicle’s trajectory a transition to the alternative in the HFSM. As for an
within this zone. Meanwhile, the driver can feel the resist- acknowledgement, the pressed button will become blue
ance on the steering wheel. (lasting 2 s). If the user presses when the green light is
off, this button will temporally become red to signify a
refusal.
4.3 HMI design
Thanks to the haptic shared control scheme at the opera-
In addition to the interaction logic, HMI plays a crucial role tional level, the driver can directly feel system’s activity
in efficient cooperation between the driver and the system. and the potential dangers through the controller’s steering
In this use case, we designed and prototyped a set of HMI. torque at the steering wheel. As a justification of the poten-
For the cooperation at the tactical level, we update the tial resistance at the steering wheel due to the presence of a
maneuver cooperation interface previously designed in (Guo traffic vehicle in the vicinity in an adjacent lane, a warning
et al. 2017a) to integrate the display and command for lane indicator like in the blind spot warning system is activated
change maneuver. The prototyped interface consists of three in the corresponding side view mirror.
elements (Fig. 6):
4.4 Preliminary results
• Highway merging context: this HMI graphic represents a
highway merging scene with the ego vehicle (blue vehi- These two cooperation methods have been separately
cle) and the merging vehicle (abstracted yellow bloc). In evaluated in our previous work. In (Guo et al. 2017a), we
this representation, the intended maneuver of the system evaluated the maneuver cooperation method via a 22-user
is symbolized by three blue arrows and available alterna- study on a driving simulator. The results suggest that the
tives are symbolized by a single arrow. Considering that system yielded higher performance in a highway merging
the driver may not supervise the driving in automated situation with cooperation than when it handled the situa-
driving mode, a sound is transmitted when the system tion alone. In terms of cooperation performance, users had
displays the highway merging context. a less error rate in average when they had the possibility to
• Simulated augmented reality: when the system detects choose any of the alternatives than when they could only
the merging vehicle, a yellow semi-transparent rectangle choose an alternative other than the system’s intention.
tracking the merging vehicle appears to inform the driver This is mainly because users tend to give a confirmation if
that the system is handling this vehicle. they share the same intention with the system, whereas this

13
Cognition, Technology & Work (2019) 21:657–670 667

confirmation has no effect on system’s behavior accord- 5 Discussion and conclusion


ing to the current interaction logic. Concerning the HMI
design, the shared representation concept helped users to This paper presented a hierarchical cooperative control
understand system’s intention, however, dispatched HMI architecture that extends existing architectures of auto-
components increased the user’s cognitive demand. mated driving systems towards cooperation between a
The MPC-based shared steering control—a cooperation driver and an automated driving system. To derive this
method at the operational level was evaluated in a simula- architecture, we first decomposed driver–vehicle coop-
tor study with 12 participants (Guo et al. 2018). The test eration using the tri-level hierarchy proposed by Michon.
results show that the shared control scheme allowed users At each level, we implemented either human supervisory
to take over the control with ease and at the same time control or shared control to resolve interferences between
ensured a smooth transition. Subjective evaluation based driver’s control and system’s control. We not only identi-
on questionnaire suggests that users in general felt safe, fied functions involved in each level’s cooperation process,
comfortable and in control when they steered simultane- but also provided new approaches to achieve human super-
ously with the system. However, a large variance of steer- visory control and shared control based on these func-
ing wheel reversals among the participants was observed tions. In the case study part, we showed how to design a
when the system exerted control to assist participants in cooperative control system with HMIs based on this gen-
lane change maneuvers, which suggested that some par- eral architecture within a use case. We also reported the
ticipants tended to correct the trajectory during the lane preliminary results of the current design and discussed the
change. issues exposed by user tests.
These user studies do not only reveal the utility of such This study demonstrated the correspondence between
cooperation methods for future users of automated driving technical concepts of automated driving and theoretical
systems, but also demonstrate the strength of this hierar- models of human cognition and control. This correspond-
chical architecture in supporting the design and analysis ence constitutes a source of new ideas and methods to
of driver–vehicle cooperation for automated driving. First, design driver–vehicle cooperation. In this work, the cor-
the decomposition of cooperation allows one to focus on respondence between the hierarchy of the layered archi-
the processes involved in the cooperation occurring at tectures of automated driving system and that of the driv-
one level while ignoring processes at other levels, thus ing task lays a basis for our architecture. We also showed
facilitating the analysis of such a complex human–machine that maneuver planning process on the system side can be
system. For example, test results on the maneuver coop- described by the model proposed by Johannsen and Rouse.
eration method imply that the system needs to initiate the Note that the “situated” concept of the behavior-based par-
cooperation with a sufficient lead time so that the driver adigm which is widely applied for robotics can also find
has time to assess the situation and to participate in the its equivalence in the field of human–machine interaction
cooperation. This observation is coherent with the assump- (refer to the hypothesis of “situated action” proposed by
tion that decisions on maneuver take place in seconds at Lucy Suchman (1987) to explain human cognitive pro-
the tactical level given by Michon’s driver model. Thus, cess). Lastly, the concept of “safe travel field” proposed
we can focus on the situation assessment function and the by Gibson and Crooks to describe human driving behavior
HMI at the tactical level to ensure a long enough time corresponds to the formulation of a constrained optimal
window for driver to cooperate while not bothering with control problem from the perspective of automatic control.
the shared control function. Second, viewing this archi- This correspondence results in the proposed MPC-based
tecture as a whole, one can study interactions between the shared control framework.
processes at different levels while ignoring the details of The scope of the current architecture was limited to
subsystems at a level. In the MPC-based shared steering achieving driver–vehicle cooperation within the system
control framework, when the system adapts its maneuver limits. In other words, the cooperative control methods
plan to that of the driver at the tactical level, it is crucial proposed in this architecture address the interference
to ensure a smooth torque change rendered by the sys- between the driver and the system in driving tactics and
tem to inform the driver of the change of system’s plan. behaviors and concern only system normal use. However,
Otherwise, an intrusive torque on the steering wheel risks driver–vehicle cooperation should also cover the usage at
frightening the user rather than reassuring him/her that the and beyond exceeding system limits. How to extend this
system intends to cooperate with him/her. cooperative control architecture to those critical situations
In the future work, in addition to addressing the issues in which the system reaches its limit is one of the axes of
exposed in the previous user tests, it is of interest to com- our future work. Another aspect that is not addressed by
pare these two cooperation methods in terms of, e.g., user’s the current architecture is the final authority issue. If the
performance and preference in a single user study.

13

668 Cognition, Technology & Work (2019) 21:657–670

system is allowed to resist the driver to prevent potential Broggi A, Cerri P, Debattisti S et al (2015) PROUD–public road
danger at the operational level, the two actors may enter urban driverless-car test. IEEE Trans Intell Transp Syst
16:3508–3519. https​://doi.org/10.1109/TITS.2015.24775​56
into conflict situations and a fundamental question needs Buehler M, Iagnemma K, Singh S (2009) The DARPA urban chal-
to be answered: who has the final authority? If this ques- lenge: autonomous vehicles in city traffic, 1st edn. Springer,
tion has not yet reached a consensus from an engineering Berlin
perspective (refer to the two contrast designing philoso- de Winter JCF, Dodou D (2011) Preparing drivers for dangerous situ-
ations: a critical reflection on continuous shared control. In: 2011
phies of Boeing and Airbus for their auto-pilot systems IEEE international conference on systems, man, and cybernetics
(Billings 1997), human factors deliberation provides new (SMC). pp 1050–1056
insights to address this question. Flemisch et al. (2012) Dickmanns ED (1986) Computer Vision in Road Vehicles—Chances
highlighted the tight relation between the “ability” and and Problems. In: ITCS-Symposium on Human Factors Technol-
ogy for Next-Generation Transportation Vehicles. pp. 16–20
the “authority”, i.e., the distribution of authority between Endsley MR, Kiris EO (1995) The out-of-the-loop performance prob-
human and machine should be based on the ability of each lem and level of control in automation. Hum Factors J Hum Fac-
agent. Therefore, it is interest to explore the distribution of tors Ergon Soc 37:381–394. https​://doi.org/10.1518/00187​20957​
the authority perhaps at a new higher level in the hierar- 79064​555
Erlien SM, Fujita S, Gerdes JC (2016) Shared steering control using
chical cooperative control architecture in the future work, safe envelopes for obstacle avoidance and vehicle stability. IEEE
e.g., using driver monitoring techniques to estimate the Trans Intell Transp Syst 17:441–451. https​://doi.org/10.1109/
driver’s current ability to handle the situation. TITS.2015.24534​04
Färber B (2016) Communication and communication problems
between autonomous vehicles and human drivers. In: Maurer
M, Gerdes JC, Lenz B, Winner H (eds) Autonomous Driving.
Springer, Berlin, pp 125–144
References Flemisch F, Heesen M, Kelsch J, Schindler J, Preusche C, Dittrich J
(2010) Shared and cooperative movement control of intelligent
Abbink DA, Mulder M (2010) Neuromuscular analysis as a guideline technical systems: Sketch of the design space of haptic-multi-
in designing shared control. In: Hosseini M (ed) Advances in Hap- modal coupling between operator, co-automation, base system and
tics. InTech, London environment. In: The 11th IFAC/IFIP/IFORS/IEA Symposium on
Abbink DA, Mulder M, Boer ER (2012) Haptic shared control: Analysis, Design, and Evaluation of Human-Machine Systems,
smoothly shifting control authority? Cogn Technol Work 14:19– Valenciennes, France
28. https​://doi.org/10.1007/s1011​1-011-0192-5 Flemisch F, Heesen M, Hesse T et al (2012) Towards a dynamic bal-
Albert M, Lange A, Schmidt A et al (2015) Assessment of interaction ance between humans and automation: authority, ability, respon-
concepts under real driving conditions. Procedia Manuf 3:2832– sibility and control in shared and cooperative control situations.
2839. https​://doi.org/10.1016/j.promf​g.2015.07.767 Cogn Technol Work 14:3–18. https​://doi.org/10.1007/s1011​
Anderson SJ, Karumanchi SB, Iagnemma K, Walker JM (2013) The 1-011-0191-6
intelligent copilot: a constraint-based approach to shared-adaptive Flemisch F, Bengler K, Bubb H et al (2014) Towards cooperative
control of ground vehicles. IEEE Intell Transp Syst Mag 5:45–54. guidance and control of highly automated vehicles: H-Mode
https​://doi.org/10.1109/MITS.2013.22477​96 and Conduct-by-Wire. Ergonomics 57:343–360. https​: //doi.
Bainbridge L (1983) Ironies of Automation. Automatica 19:775–779. org/10.1080/00140​139.2013.86935​5
https​://doi.org/10.1016/0005-1098(83)90046​-8 Flemisch F, Abbink D, Itoh M et al (2016) Shared control is the sharp
Baltzer M, Flemisch F (2017) Arbitration: When Time for Interaction end of cooperation: towards a common framework of joint action,
Mediation in a Human Machine System is running out. Cogni shared control and human machine cooperation. IFAC-Pap 49:72–
Technol Work, Special Issue “Shared and cooperative control of 77. https​://doi.org/10.1016/j.ifaco​l.2016.10.464
safety critical systems” (in preparation) Geyer S (2013) Maneuver-based vehicle guidance based on the con-
Bauer E, Lotz F, Pfromm M et al (2012) PRORETA 3: an integrated duct-by-wire principle. In: Maurer M, Winner H (eds) Automotive
approach to collision avoidance and vehicle automation. Automa- systems engineering. Springer, Berlin Heidelberg, pp 111–132
tisierungstechnik 60:755–765 Gibson JJ, Crooks LE (1938) A theoretical field-analysis of automobile-
Bender P, Taş ÖŞ, Ziegler J, Stiller C (2015) The combinatorial aspect driving. Am J Psychol 51:453–471. https:​ //doi.org/10.2307/14161​
of motion planning: Maneuver variants in structured environ- 45
ments. In: 2015 IEEE Intelligent Vehicles Symposium (IV). pp Glaser S, Vanholme B, Mammar S, Gruyer D, Nouvelière L (2010)
1386–1392 Maneuver-based trajectory planning for highly autonomous
Benloucif MA, Nguyen AT, Sentouh C, Popieul JC (2017) A new vehicles on real road with traffic and driver interaction. IEEE
scheme for haptic shared lateral control in highway driving using Trans Intell Transp Syst 11(3):589–606. https​://doi.org/10.1109/
trajectory planning. In: 2017 IFAC World Congress. Toulouse, TITS.2010.20460​37
France. https​://doi.org/10.1016/j.ifaco​l.2017.08.2223 Gold C, Damböck D, Lorenz L, Bengler K (2013) “Take over!”
Billings CE (1997) Aviation automation: the search for a human- How long does it take to get the driver back into the loop? Proc
centered approach. Lawrence Erlbaum Associates Publishers, Hum Factors Ergon Soc Annu Meet 57:1938–1942. https​://doi.
Mahwah org/10.1177/15419​31213​57143​3
Blanco M, Atwood J, Vasquez H, et al (2015) Human factors evalu- Gu T, Dolan JM, Lee JW (2016) Automated tactical maneuver discov-
ation of Level 2 and Level 3 automated driving concepts. Tech- ery, reasoning and trajectory planning for autonomous driving. In:
nical report. Washington, DC: National Highway Traffic Safety 2016 IEEE/RSJ International Conference on Intelligent Robots
Administration and Systems (IROS). pp 5474–5480
Brandt T (2008) A predictive potential field concept for shared vehicle Guo C, Sentouh C, Popieul JC, et al (2015) Shared control frame-
guidance. Dissertation, Universität Paderborn work applied for vehicle longitudinal control in highway

13
Cognition, Technology & Work (2019) 21:657–670 669

merging scenarios. In: 2015 IEEE International Conference on Michon JA (1985) A critical view of driver behavior models: what
Systems, Man, and Cybernetics (SMC). pp 3098–3103 do we know, what should we do? In: Evans L, Schwing RC (eds)
Guo C, Sentouh C, Soualmi B, et al (2016) Adaptive Vehicle Longi- Human behavior and traffic safety. Springer, US, pp 485–524
tudinal Trajectory Prediction for Automated Highway Driving. Mitschke M, Wallentowitz M (2004) Dynamik der Kraftfahrzeuge,
In: 2016 IEEE Intelligent Vehicles Symposium (IV) 4th edn. Springer, Berlin
Guo C, Sentouh C, Popieul JC et al (2017a) Cooperation between Mulder M, Abbink DA, van Paassen MM, Mulder M (2011) Design
driver and automated driving system: Implementation and eval- of a haptic gas pedal for active car-following support. IEEE
uation. Transp Res Part F Traffic Psychol Behav. 5:6. https​://doi. Trans Intell Transp Syst 12:268–279. https​://doi.org/10.1109/
org/10.1016/j.trf.2017.04.006 TITS.2010.20914​07
Guo C, Sentouh C, Popieul JC, Haué JB (2017b) MPC-based shared Nguyen AT, Sentouh C, Popieul JC (2016) Driver-Automation Coop-
steering control for automated driving systems. In: 2017 IEEE erative Approach for Shared Steering Control under Multiple
International conference on systems, man, and cybernetics System Constraints: Design and Experiments. IEEE Trans Ind
(SMC). https​://doi.org/10.1109/smc.2017.81225​90 Electron. https​://doi.org/10.1109/tie.2016.26451​46
Guo C, Sentouh C, Popieul JC, Haue JB (2018) Predictive shared Pacaux-Lemoine M-P, Flemisch F (2016) Layers of shared and coop-
steering control for driver override in automated driving: a erative control, assistance and automation. IFAC-Pap 49:159–
simulator study. Transp Res Part F Traffic Psychol Behav. https​ 164. https​://doi.org/10.1016/j.ifaco​l.2016.10.479
://doi.org/10.1016/j.trf.2017.12.005 Park J, Karumanchi S, Iagnemma K (2015) Homotopy-based divide-
Hidas P (2005) Modelling vehicle interactions in microscopic simu- and-conquer strategy for optimal trajectory planning via mixed-
lation of merging and weaving. Transp Res Part C Emerg Tech- integer programming. IEEE Trans Robot 31:1101–1115. https​
nol 13:37–62. https​://doi.org/10.1016/j.trc.2004.12.003 ://doi.org/10.1109/TRO.2015.24593​73
Hoc J-M (2001) Towards a cognitive approach to human–machine Petermeijer SM, Abbink DA, de Winter JCF (2015) Should drivers
cooperation in dynamic situations. Int J Hum-Comput Stud be operating within an automation-free bandwidth? Evaluating
54:509–540. https​://doi.org/10.1006/ijhc.2000.0454 haptic steering support systems with different levels of author-
Hoc J-M, Cacciabue PC, Hollnagel E, Cacciabue PC (eds) (1994) ity. Hum Factors 57:5–20. https​://doi.org/10.1177/00187​20814​
Expertise and technology: cognition and human-computer coop- 56360​2
eration. Psychology Press, Hillsdale Pirjanian P (1999) Behavior Coordination Mechanisms—State-of-the-
Hoc J-M, Young MS, Blosseville J-M (2009) Cooperation between Art. Report (No. IRIS-99–375). University of Southern California,
drivers and automation: implications for safety. Theor Issues Institute for Robotics and Intelligent Systems
Ergon Sci 10:135–160. https​: //doi.org/10.1080/14639​2 2080​ Quain JR (2016) How to use tesla’s autopilot (and how not to). In:
23688​56 Toms guide. http://www.tomsg​uide.com/us/how-to-use-tesla​
Itoh M, Inagaki T (2014) Design and evaluation of steering protec- -autop​ilot,revie​w-3870.html. Accessed 9 Feb 2017
tion for avoiding collisions during a lane change. Ergonomics SAE (2016) Taxonomy and definitions for terms related to driv-
57:361–373. https​://doi.org/10.1080/00140​139.2013.84847​4 ing automation systems for on-road motor vehicles. Standard
Johannsen G, Rouse WB (1979) mathematical concepts for mod- J3016_201609. SAE International
eling human behavior in complex man-machine systems. Hum Saleh L, Chevrel P, Claveau F et al (2013) Shared steering control
Factors J Hum Factors Ergon Soc 21:733–747. https​: //doi. between a driver and an automation: stability in the presence
org/10.1177/00187​20879​12210​610 of driver behavior uncertainty. IEEE Trans Intell Transp Syst
Kelsch J, Flemisch F, Löper C, Schieben A, Schindler J (2006) Links 14:974–983. https​://doi.org/10.1109/TITS.2013.22483​63
oder rechts, schneller oder langsamer? Grundlegende Frages- Sentouh C, Soualmi B, Popieul JC, Debernard S (2013) Cooperative
tellungen beim cognitive systems engineering von hochau- steering assist control system. In: 2013 IEEE international confer-
tomatisierter Fahrzeugführung. In: DGLR Fachauschusssit- ence on systems, man, and cybernetics. pp 941–946
zung Anthropotechnik “Cognitive Systems Engineering in der Sheridan TB (1992) Telerobotics, automation, and human supervisory
Fahrzeug- und Prozessführung”, Karlsruhe, Germany control. MIT Press, Cambridge
Kortenkamp D, Simmons R (2008) Robotic Systems Architectures Shladover SE, Desoer CA, Hedrick JA et al (1991) Automated vehicle
and Programming. In: Prof BS, Prof OK (eds) Springer hand- control developments in the PATH program. IEEE Trans Veh Tech
book of robotics. Springer, Berlin Heidelberg, pp 187–206 40(1):114–130. https​://doi.org/10.1109/25.69979​
Kumagai T (2006) Prediction of human driving behavior using Soualmi B, Sentouh C, Popieul JC, Debernard S (2014) Automation-
dynamic bayesian networks. IEICE Trans Inf Syst E89:857–860. driver cooperative driving in presence of undetected obstacles.
https​://doi.org/10.1093/ietis​y/e89-d.2.857 Control Eng Pract 24:106–119. https​://doi.org/10.1016/j.conen​
Löper C, Flemisch F (2009) Ein Baustein für hochautomatisiertes gprac​.2013.11.015
Fahren: Kooperative, manöverbasierte Automation in den Pro- Suchman LA (1987) Plans and situated actions: the problem of human-
jekten H-Mode und HAVEit. In: 6th Workshop Fahrerassisten- machine communication. Cambridge University Press, Cambridge
zsystem, Löwenstein Germany Takada Y, Boer ER, Sawaragi T (2013) Driving assist system: shared
Lorenz L, Kerschbaum P, Schumann J (2014) Designing take over haptic human system interaction. IFAC Proc 46:203–210
scenarios for automated driving How does augmented real- Taş ÖŞ, Kuhnt F, Zöllner JM, Stiller C (2016) Functional system archi-
ity support the driver to get back into the loop? Proc Hum tectures towards fully automated driving. In: 2016 IEEE Intel-
Factors Ergon Soc Annu Meet 58:1681–1685. https​: //doi. ligent vehicles symposium (IV). pp 304–309
org/10.1177/15419​31214​58135​1 Ulbrich S, Maurer M (2015) Situation assessment in tactical lane
Markoff J (2010) Google cars drive themselves, in traffic. N. change behavior planning for automated vehicles. In: 2015 IEEE
Y. Times. Accessed July 05, 2017. http://www.nytim​ 18th International conference on intelligent transportation sys-
es.com/2010/10/10/scien​ce/10goo​gle.html tems. pp 975–981
Matarić MJ, Michaud F (2008) Behavior-Based Systems. In: Prof Walch M, Lange K, Baumann M, Weber M (2015) Autonomous driv-
BS, Prof OK (eds) Springer handbook of robotics. Springer, ing: investigating the feasibility of car-driver handover assistance.
Berlin Heidelberg, pp 891–909 In: Proceedings of the 7th international conference on automotive
user interfaces and interactive vehicular applications. ACM, New
York, NY, USA, pp 11–18

13

670 Cognition, Technology & Work (2019) 21:657–670

Wetmore JM (2003) Driving the dream-the history and motivations International Conference on Intelligent Transportation Systems
behind 60 years of automated highway systems in America. Auto (ITSC). pp 13–18
History Rev 7:4–19
Willmore TJ (2012) An introduction to differential geometry, Reprint Publisher’s Note Springer Nature remains neutral with regard to
edn. Dover Publications, New York jurisdictional claims in published maps and institutional affiliations.
Wolf I (2016) The interaction between humans and autonomous agents.
Autonomous driving. Springer, Berlin, pp 103–124
Xiong Z, Dixit VV, Waller ST (2016) The development of an Ontology
for driving Context Modelling and reasoning. In: 2016 IEEE 19th

13

You might also like