Search

International audience; The availability of large amounts of medical data, and advances in data science, make artificial intelligence (AI) algorithms the most targeted tool to address the challenge of personalized clinical decision-making for complex diseases. These algorithms exceed the ability of individuals to predict future events in real time from existing data, if the dataset is complete and reflects the entire target population, and the algorithms provide reliable prediction models. However, the process of moving from the data inputs to output decisions is not obvious to a physician. This “black box” problem leads to discomfort, concerns and controversies about the reliability of these models, which limits their adoption [1]. This study aims to enhance the prediction of complex diseases course and promote the adoption of AI-based medical systems. We propose a new approach based on the development of hybrid predictive models leveraging computational power, good quality medical data and interaction with physicians, especially their intuitive reasoning. This approach integrates the physician in the prediction model creation process, especially in the data preprocessing phase for training the algorithm. Rather than using the raw values of the patient’s features and performing auto-clustering by the algorithm, we propose to use a clustering based on the physician’s reasoning. Nevertheless, for acceptability reasons, a deep understanding of medical practice is essential to design decision support tools. We have initiated the application of this new approach to multiple sclerosis (MS) [2]. MS is a chronic disease that affects the central nervous system with a multifactorial pathophysiology and heterogeneous course. Through interactions with MS physicians (interviews, observations) and the literature [3–5], we have identified relevant factors controlling MS course. We converted the factors’ values in patients’ data from a clinical trial dataset (NCT00906399) based on the physician’s reasoning. For example, according to the physician, there are three categories of patients for the factor ‘disease duration’ [D1: 10 years]. For a patient with 15 years of disease, instead of using the raw value, we use the equivalent group (D3) according to the physician’s reasoning. We apply the same data transformation technique for all available factors used by expert physicians and validated in the literature. The converted data are then used as input for the AI algorithm to create a predictive model for MS course. Once the model is validated, we intend to use interactive visuals easily understandable and providing a synchronous communication between the algorithm and the physician. Thus, the latter can adjust the factors (change the type of treatment to propose to the patient, the lifestyle to advise, etc.) until the results reflect her/his medical intent. This new approach based on physicians’ reasoning will increase physician confidence in AI models, and will also contribute to improve the quality of prediction results, as learning data is included in the decision-making process of the medical practice. The use of interactive and easy-to-understand visualizations will facilitate physician-algorithm communication [6] and help to better understand the results provided by the AI model [7]. Hence, boosting the adoption of AI-based medical systems. An implementation of this new approach is in progress to support medical decision-making in multiple sclerosis. In the future, we expect to compare the results of the prediction based on the physicians’ factors with the prediction based on all existing factors in the database. If some factors are found to be important, we will use them in conjunction with the physicians’ factors.

Machine learning (ML) models have proven their potential in acquiring and analyzing large amounts of data to help solve real-world, complex problems. Their use in healthcare is expected to help physicians make diagnoses, prognoses, treatment decisions, and disease outcome predictions. However, ML solutions are not currently deployed in most healthcare systems. One of the main reasons for this is the provenance, transparency, and clinical utility of the training data. Physicians reject ML solutions if they are not at least based on accurate data and do not clearly include the decision-making process used in clinical practice. In this paper, we present a hybrid human–machine intelligence method to create predictive models driven by clinical practice. We promote the use of quality-approved data and the inclusion of physician reasoning in the ML process. Instead of training the ML algorithms on the given data to create predictive models (conventional method), we propose to pre-categorize the data according to the expert physicians’ knowledge and experience. Comparing the results of the conventional method of ML learning versus the hybrid physician–algorithm method showed that the models based on the latter can perform better. Physicians’ engagement is the most promising condition for the safe and innovative use of ML in healthcare.

International audience; Haptic shared control has proven to be an effective method to assist a driver in controlling a vehicle. This method is now being considered for use in developing strategies for smooth transitions between manual and autonomous driving modes. This paper has two objectives. First, it proposes to adapt an existing haptic shared control strategy to achieve transitions between manual and autonomous modes and to evaluate this approach with real drivers on a driving simulator. Second, it proposes to evaluate four different transition profiles in an obstacle avoidance context. The first profile is a gradual transition from autonomous mode to shared control mode, followed by another transition from shared control mode to autonomous mode once the obstacle is passed. The second is a gradual transition from autonomous mode to manual mode. The third is a binary transition from autonomous mode to manual mode. Finally, in the fourth condition, the driver overrides the autonomous mode. These transition profiles were evaluated in curves and straight lines on a driving simulator. The results first validated the use of the haptic shared control strategy to execute transitions between manual and autonomous modes. The distribution of the torques delivered by the automation system and the driver corresponded to the progression of the expected sharing level. Second, the gradual transitions showed advantages over binary transitions and the override of the autonomous mode, both in terms of steering performance and subjective evaluation.

International audience; Autonomous navigation becomes complex when it is performed in an environment that lacks road signs and includes a variety of users, including vulnerable pedestrians. This article deals with the perception of collision risk from the viewpoint of a passenger sitting in the driver’s seat who has delegated the total control of their vehicle to an autonomous system. The proposed study is based on an experiment that used a fixed-base driving simulator. The study was conducted using a group of 20 volunteer participants. Scenarios were developed to simulate avoidance manoeuvres that involved pedestrians walking at 4.5 kph and an autonomous vehicle that was otherwise driving in a straight line at 30 kph. The main objective was to compare two systems of risk perception: These included subjective risk assessments obtained with an analogue handset provided to the participants and electrodermal activity (EDA) that was measured using skin conductance sensors. The relationship between these two types of measures, which possibly relates to the two systems of risk perception, is not unequivocally described in the literature. This experiment addresses this relationship by manipulating two factors: The time-to-collision (TTC) at the initiation of a pedestrian avoidance manoeuvre and the lateral offset left between a vehicle and a pedestrian. These manipulations of vehicle dynamics made it possible to simulate different safety margins regarding pedestrians during avoidance manoeuvres. The conditional dependencies between the two systems and the manipulated factors were studied using hybrid Bayesian networks. This relationship was inferred by selecting the best Bayesian network structure based on the Bayesian information criterion. The results demonstrate that the reduction of safety margins increases risk perception according to both types of indicators. However, the increase in subjective risk is more pronounced than the physiological response. While the indicators cannot be considered redundant, data modeling suggests that the two risk perception systems are not independent.

International audience

21st Congress of the International Ergonomics Association (IEA 2021), VANCOUVER, CANADA, 14-/06/2021 - 18/06/2021; International audience; Using an integrative user-centered approach, the research project aimed at designing and assessing an adaptive HMI to improve the safety and the quality of the take-over in a level-3 automated vehicle. Future autonomous driving users confirmed the need for the monitoring systems to improve safety as long as they are simple and not intrusive. They request specific information to understand and initiate appropriated actions in case of a critical situation. However, to avoid overloading and detri-mental effects, it is important that the system limits the amount of infor-mation at the right moment.

When manually steering a car, the driver’s visual perception of the driving scene and his or her motor actions to control the vehicle are closely linked. Since motor behaviour is no longer required in an automated vehicle, the sampling of the visual scene is affected. Autonomous driving typically results in less gaze being directed towards the road centre and a broader exploration of the driving scene, compared to manual driving. To examine the corollary of this situation, this study estimated the state of automation (manual or automated) on the basis of gaze behaviour. To do so, models based on partial least square regressions were computed by considering the gaze behaviour in multiple ways, using static indicators (percentage of time spent gazing at 13 areas of interests), dynamic indicators (transition matrices between areas) or both together. Analysis of the quality of predictions for the different models showed that the best result was obtained by considering both static and dynamic indicators. However, gaze dynamics played the most important role in distinguishing between manual and automated driving. This study may be relevant to the issue of driver monitoring in autonomous vehicles.

International audience; During highly automated driving, drivers no longer physically control the vehicle but they might need to monitor the driving scene. This is true for SAE level 2, where monitoring the external environment is required; it is also true for level 3, where drivers must react quickly and safely to a take-over request. Without such monitoring, even if only partial, drivers are considered out-of-the-loop (OOTL) and safety may be compromised. The OOTL phenomenon may be particularly important for long automated driving periods during which mind wandering can occur. This study scrutinized drivers’ visual behaviour for 18 min of highly automated driving. Intersections between gaze and 13 areas of interest (AOIs) were analysed, considering both static and dynamic indicators. An estimation of self-reported mind wandering based on gaze behaviour was performed using partial least squares (PLS) regression models. The outputs of the PLS regressions allowed defining visual strategies associated with good monitoring of the driving scene. This information may enable online estimation of the OOTL phenomenon based on a driver’s spontaneous visual behaviour.

International audience; Driver modeling is essential in the development of haptic guidance systems. A new cybernetic driver model designed to account for the cooperation between the driver and haptic guidance systems has recently been proposed. This paper aims to validate this model in situations of interaction with different levels of haptic guidance on a driving simulator. Two experiments have been performed for this purpose. The first experiment consisted of implementing the driver model in the driving simulator and evaluating its lateral control performance when interacting with a haptic guidance system. The results reveal that the model can be adapted to different sharing levels by adjusting only the gain of an internal model of the steering wheel compliance. The second experiment consisted of estimating the evolution of the gain of this internal model using the unscented Kalman filter. The results reveal consistency between the evolution of the identified parameter and the level of sharing of the haptic guidance system. The driver model represents the process of human driver adaptation to variations in the level of sharing in haptic guidance systems.

International audience; In autonomous cars, the automation systems assume complete operational control. In this situation, it is essential that passengers always feel comfort-able with the vehicle's decisions. In this project, we are specifically inter-ested in risk assessment by the passenger of an autonomous car navigating among pedestrians in a shared space. A driving simulator experiment was conducted with 27 participants. The challenge was twofold: on the one hand, to find a link between the pedestrians' avoidance behavior of the ve-hicle and the risk felt by the passenger; and on the other hand, to try to pre-dict this perceived risk in real time. The study revealed a significant effect of two factors on the risk assessed by the participants: (1) the value of the TTC at the moment the vehicle begins a pedestrian avoidance maneuver; (2) the lateral distance it leaves to the pedestrian. The proposed real-time pre-diction model is based on the principle of impulse response operation. This new paradigm assumes that the passenger's risk assessment is the result of a quantifiable unconscious internal phenomenon that has been estimated us-ing the dynamics of the perceived pedestrian approach. The results showed that this approach was predictive of risk for isolated avoidance maneuvers, but was insufficient to explain the variability in the risk assessment behav-ior of the participants.

International audience; When sharing virtual collaborative environments, operators exchange spatial statements that refer to the objects’ positions in the virtual space. If operators are to understand each other, they need to develop a common spatial frame of reference and then choose a space coding to describe the objects’ positions. In this paper, we consider how the content of a virtual environment can influence communication between users. We designed two studies in which one participant (the speaker) had to indicate the position of one object to another participant (the addressee). The virtual environment was sometimes enriched by additional (proximal and distal) visual cues. In study 1, we considered statements production. We observed that the speakers most often used the avatar of their partner as a spatial reference to indicate a localization in the virtual space (i.e., Addressee-Centered coding) despite it increases their mental workload. Nevertheless, in complex situations, they also used distal cues to speak to the addressees (i.e., Exocentric coding of the space). In study 2, we considered statements comprehension. Addressee-Centered coding and Exocentric coding were used by the speakers in various spatial configurations to indicate the object position. We observed that Exocentric coding is the most difficult to manage for the addressee. These results indicate that speakers implemented the principle of less collaborative effort by adopting a way of exchanging information based on an asymmetrical cognitive cost, taking into consideration each other’s difficulties. This allows a balanced mental workload to be maintained between the two operators throughout the task.

International audience

International audience; Advanced Driver-Assistance Systems (ADAS) have become an essential part of modern cars. Among the solutions proposed, haptic shared control of the steering wheel is increasingly being studied. A fundamental question is how drivers adapt their behavior to these systems. This article proposes to use the Unscented Kalman Filter (UKF) to identify the variation over time in the psychological and neuromuscular parameters of a driver structured model. The goal here is to understand how the driver adapts to changes, whether regarding the behavior of the steering system, the visibility or the road conditions. The LPV system considered for identification is known as the cybernetic driver model. Two experiments carried out respectively with Simulink© and on a driving simulator provide the data. The methodology proposed for tuning the UKF is studied from the results obtained with those data. A multi-UKF strategy is also considered. The methodology reveals useful when a compromise between rapidity and precision has to be achieved for parameters estimation. It opens the way to a detailed analysis of the driver's parameter variations within the multi-UKF framework.

International audience; Haptic shared control is a consistent way to design an assistance for the lateral control of a vehicle. The most important problem raised by haptic shared control is to minimize useless conflicts between the driver and the assistance. To deal with, this paper proceeds in two stages. The first is concerned with feedforward synthesis in a new way, by identifying mainly the geometric part of driving from real driver data. The torque to apply to the steering wheel and more generally the reference trajectory is thus obtained from a second order model fed by the road curvature. Then, the feedback is designed based on a driver-vehicle-road model, using a mixed í µí°» 2 /í µí°» ∞ control synthesis involving both lane following performance and sharing capabilities indicators. Finally, the shared control strategy is simulated on Matlab/Simulink using a vehicle-road model. The results obtained show good features both in terms of lane following and sharing performances. This control strategy seems then to be an interesting candidate for haptic shared control.

This paper reviews a set of scientific studies on how driver modelling may serve as the basis for designing advanced driving assistance systems. The work was aimed at explicitly representing the human visual and motor processes involved in the control of steering, and took into account current knowledge in the behavioural sciences. The nature and structure of the model, and its calibration using experimental data (identification), were addressed. Two design applications were considered: 1) estimating the driver state in various conditions of distraction and 2) building an automatic controller for haptic shared control of the steering wheel.

International audience

International audience; Within the context of more and more autonomous vehicles, an automatic lateral control device (AS: Automatic Steering) was used to steer the vehicle along the road without drivers’ intervention. The device was not able to detect and avoid obstacles. The experiment aimed to analyse unexpected obstacle avoidance manoeuvres when lateral control was delegated to automation. It was hypothesized that drivers skirting behaviours and eye movement patterns would be modified with automated steering compared with a control situation without automation. Eighteen participants took part in a driving simulator study. Steering behaviours and eye movements were analysed during obstacle avoidance episodes. Compared with driving without automation, skirting around obstacles was found to be less effective when drivers had to return from automatic steering to manual control. Eye movements were modified in the presence of automatic steering, revealing further ahead visual scanning of the driving environment. Resuming manual control is not only a problem of action performance but is also related to the reorganisation of drivers’ visual strategies linked to drivers’ disengagement from the steering task. Assistance designers should pay particular attention to potential changes in drivers’ activity when carrying out development work on highly automated vehicles.

Aims Lane departures while driving a car may be caused directly by errors in steering behaviour and indirectly as a consequence of driver distraction. Designing advanced driving assistance systems (ADAS) is a way to avoid these situations. One of the key problems is to monitor and predict driver behaviour in order to make the ADAS intervene in a timely and efficient way. Our approach to achieve this goal is to incorporate a driver model in the design process. The model represents the visual and motor determinants of steering control. It has been used as the foundation of two types of ADAS: 1. a haptic shared control system that exerts forces on the steering wheel in such a way that the automation blends into the driver’s sensorimotor control loop, providing continuous support to lane keeping, and 2. a method of distraction estimation based on the comparison between steering behaviour and the model prediction. Methods Based on current knowledge of human sensorimotor functions, the model represents the visual anticipation of the road curvature fed by the angular deviation of a far point, the visual compensation of lateral position error fed by the angular deviation of a near point, and a neuromuscular system that transforms the output of the visual subsystem into a steering wheel torque. For haptic shared control, the approach consisted in designing a control law that optimizes performance and cooperation criteria, taking into account the predictions of the driver model. A comparison was performed between the driver behaviour when using this system and when using a shared control law that did not incorporate a driver model. For distraction estimation, an identification of the driver model’s parameters was conducted in five different distraction conditions: no distraction and cognitive, visual, motor and visuomotor distractions. The torque prediction error and the parameters values were analysed. Results The results concerning haptic shared control suggest that cooperation between the driver and the automation can be improved by the inclusion of the driver model in the control law. In particular, when the level of haptic authority was set at a high level, the driver acted 27% more in coherence with the system. For the second application, the results show that the model prediction error is commensurate to a modification of steering behaviour caused by distraction. The parametric analysis suggests that an online identification of the model may be used to some extent to discriminate between different types of distraction. Conclusions Adopting a model-based approach to design ADAS is promising. The current model is limited to the representation of sensorimotor processes in humans. As such, it is adapted to be used in the context of real-time control embedded systems. In the future, it could be extended by incorporating a tactical analysis of the driving context.

International audience; The study presented here follows a former study, lead in a real aeronautical factory where operators control a semi-automated process of metallic frames stretching. The observations and interviews from this first study conducted to the identification of a misleading scheme. This misleading scheme consists in programming a stretching machine according to the results obtained on a precedent frame instead of using the inner frame characteristics. However, since studying this misleading scheme turned out to be complicated in the real factory, a micro-world, reproducing the factory activity was developed. This new article present the results obtained on this micro-world. First, the activity of twelve participants, who had never worked in the factory, was similar to the operator activity as they used the misleading scheme. Moreover, the micro-world showed that the misleading scheme brings a decrease in the performance of participants who were given more time on the micro-world. On the contrary, participant with little training always performed badly, whenever they used the misleading scheme or not. The results showed that the micro-world granted the access to a specific cognitive criterion, the misleading scheme, which was not easily accessible in the real factory. This results lead to two main recommendations: the first one states the importance to give the operator inhibition training against the misleading scheme; the second is with regard to the features an interface helping with the fabrication process should have. ACM Classification Keywords H.5.m. Information interfaces and presentation (e.g., HCI): Laboratory experiments. J.4: Social and behavioral sciences; L'étude présentée ici fait suite à une étude de terrain réalisée dans un atelier d'aéronautique où des opérateurs contrôlent un procédé semi-automatisé d'étirage de cadres métalliques. Des observations et entretiens avaient permis d'identifier un schème trompeur, qui se manifeste par l'utilisation des résultats obtenus sur un cadre afin de paramétrer une machine d'étirage pour un cadre suivant, alors que les propriétés physiques des cadres diffèrent entre eux. L'étude de ce schème s'étant révélée difficile dans l'atelier réel, un micromonde, reprenant les caractéristiques principales de l'atelier, a été développé. Ce nouvel article décrit les résultats obtenus sur ce micromonde. Le micromonde a permis de reproduire l'activité observée dans l'atelier chez douze participants tout-venants : ils ont mis en oeuvre le schème trompeur comme les opérateurs de l'atelier. De plus, dans le micromonde, le schème trompeur provoque une diminution des performances mais uniquement chez des participants ayant bénéficié de plus de temps d'entrainement sur le micromonde, les autres participants ayant dans tous les cas des résultats faibles. Les résultats montrent que l'utilisation du micromonde a permis dans ce cas d'étudier un élément cognitif spécifique, ici le schème trompeur, alors qu'il était difficile d'accès dans l'atelier. En outre, ces résultats aboutissent à deux préconisations principales : l'une concerne la nécessité d'une formation spécifique incitant les opérateurs à inhiber le schème trompeur, l'autre concerne le type d'interface à mettre en oeuvre dans un assistant au procédé de fabrication.

International audience; The goal of this study was to examine the effects of visual roll tilt on gaze and riding behavior when negotiating a bend using a motorcycle simulator. To this end, experienced motorcyclists rode along a track with a series of right and left turns whilst the degree of visual roll tilt was manipulated in three different conditions. Gaze behavior was analyzed by using the tangent point as a dynamic spatial reference; the deviation of gaze to this particular point was computed in both the horizontal and vertical directions. Steering control was assessed in terms of the lateral positioning, steering stability and number of lane departures. In the no-roll condition, the motorcyclists tracked a steering point on the road ahead, which was compatible with the hypothesis of ‘‘steer where you look” behavior. In the roll condition, our results revealed that the horizontal distribution of gaze points relative to the tangent point was preserved. However, significantly more fixations were made further ahead of the tangent point in the vertical direction. This modification of visual behavior was coupled with a degradation in steering stability and an offset in lateral positioning, which sometimes led to lane departures. These results are discussed with regard to models of visual control of steering for bend negotiation.

International audience; In the context of a remote collaboration task in virtual reality, this study aimed to analyze the effects of task distribution on the processing of spatial information and mental workload in spatial dialogs. Pairs of distant participants with specific roles (a guide and a manipulator) had to collaboratively move a virtual object in a plane factory mock-up. The displays allowed the participants to be immersed together in the virtual environment. We analyzed the dialogs that took place according to the frames of reference and the mental transformations required to produce the spatial statements. We also measured the associated mental workload. Results showed that when participants took a perspective, the manipulator’s point of view was preferred. Perspective-taking only yielded a moderate increase in mental rotations, which may explain a specifically high mental demand score for the guides’ NASA-TLX. Overall, this is in accordance with the least collaborative effort principle. This study reinforces the idea that, in collaboration, operators do not need the same aids as each other. Thus, it is not necessary to develop symmetrical tools, i.e., the same tools for all co-workers; instead, the needs of each operator should be taken into account, according to the task he has to perform. In our case, the guides would be helped with perspective-taking aids, while the manipulators would be helped with action-oriented tools.

Designing a human-machine interface for a manufacturing process requires a good knowledge of both the work domain and the operators' representation. Ecological Interface Design (EID) offers some interesting tools that can be of help in the design process. The literature on cognitive control also offers a good understanding of operators' cognitive resources. Analysing the activity of both expert and novice operators through these two frameworks may help us to better understand the differences between them. A three-step protocol was followed: 1. the elaboration of a means-end hierarchy, 2. the extraction of schemes via interviews, and 3. the evaluation of the behavioural manifestation of schemes. In the present case study, interviews revealed that both the novice and expert operators of a manufacturing process shared a representation of the global process. However, in contrast with the expert operators, the novice operators did not develop an operative scheme that related to the machine. The results will be used as a basis for the design of a human-machine interface that will aid them to do so.

National audience

International audience

International audience

International audience; Previous research has shown that a device called “motor priming” (MP) was more effective than other lane departure warning systems. MP prompts drivers to take action by means of small asymmetric oscillations of the steering wheel. The first objective of this experiment was to provide a deeper understanding of MP mechanisms through a series of comparisons with other haptic and auditory systems. The results suggest that much of the improvement in recovery manoeuvres observed with MP is due to the motor cue (proprioceptive preactivation of the gesture). Other factors, such as delivering the signal directly to the hands (stimulation of response effectors) or using the tactile modality rather than auditory warning, play a lesser role. This supports the hypothesis that MP devices directly intervene at the motor level, in contrast to more traditional warning systems, which only improve situation diagnosis. The second objective was to assess drivers' acceptance of the assistance devices. A dissociation between efficiency and acceptance of the devices was observed: drivers globally preferred auditory warning to MP. The combination of auditory warning and motor priming appeared to be a good compromise to achieve both effectiveness and acceptance. This experiment illustrates the relevance of simulator studies when dangerous situations are the main targets of the investigation.

International audience; The work described in this paper is part of a larger research program named Partage whose goal is to appropriately share lateral control between the driver and an electronic copilot or assistance. The present study is looking for a cybernetic model of a driver steering road vehicle, which makes the sensorimotor dynamics explicit. The proposed model takes into account both visual and kinesthetic perception, and includes compensatory and anticipatory processes. As such, it extends previous works. Starting from simulated car trajectories, the unknown parameters of the model are identified using the grey box identification concept. The driver-vehicle system is then analyzed with respect to the parameter sensitivity.

International audience; This study investigated the link between drivers' gaze positioning and steering behavior when negotiating bends. This was conducted by directing the driver's point of gaze toward a target situated in the vicinity of the tangent point (TP), a region known to attract a significant amount of ocular fixations and thought to provide some useful input for anticipatory steering (M. F. Land & D. N. Lee, 1994). The orientation of gaze relative to the TP was manipulated and the resulting steering behavior was compared to that obtained with a free-gaze strategy. The data revealed that constraining eye movements did not impair steering behavior. On the contrary, the continuous tracking of the fixation point promoted smoother steering control, irrespective of the position of that point. This confirms that previewing the road curvature by tracking a distant point contributes to the stability of steering. The direction of the TP does not appear to be an essential parameter in that process (D. D. Salvucci & R. Gray, 2004). The results also indicate that continuously looking at the TP or further inside the bend yielded a deviation of the trajectory. This is consistent with the hypothesis that drivers look inside the lane boundaries to determine the future path (R. M. Wilkie & J. P. Wann, 2006).

International audience; OBJECTIVE: This paper's first objective is to determine whether motor priming assistance (consisting of directional steering wheel vibrations) can be of some benefit compared with more traditional auditory (lateralized sound) or vibratory (symmetric steering wheel oscillation) warning devices. We hypothesize that warning devices favor driving situation diagnosis, whereas motor priming can improve the initiation of action even further. Another objective is to assess the possible benefits of using multimodal information by combining auditory warning with simple steering wheel vibration or motor priming. BACKGROUND: Within the context of active safety devices, the experiment dealt with moderately intrusive driving assistance devices that intervene when a certain level of risk in terms of lane departure is reached. METHOD: An analysis of the steering behavior of 20 participants following episodes of visual occlusion was carried out. Five warning and motor priming devices were compared. RESULTS: All tested devices improved the drivers' steering performance, although their effects were modulated by the drivers' risk assessment. However, performance improvements were found to be greater with a motor priming device. No additional performance enhancement was observed when auditory warning was added to steering wheel vibration or motor priming devices. CONCLUSION: This study confirms the hypothesis that the direct intervention of motor priming at the action level is more effective than a simple warning, which intervenes upstream in situation diagnosis. Multimodal information did not seem to improve driver performance. APPLICATION: This study proposes a new kind of lateral control assistance, which acts at a sensorimotor level, in contrast with traditional warning devices.

International audience; The present study investigated the influence of eye orientation upon the experience of pain. Quasi continuous electrocutaneous stimuli which slowly increased in intensity were delivered to 32 healthy females volunteers. Participants were instructed to direct the eyes at locations that were ipsilateral or contralateral to the stimulated hand. Unpleasantness threshold and pain threshold were significantly higher when the eyes were oriented ipsilateral towards the stimulated hand. In a second experiment phase, the pain intensity increased until tolerance. There was no effect of eye orientation upon pain threshold and tolerance. Results of the first experimental part are in line with the counterintuitive idea that selective monitoring reduces pain distress. The lack of significant results in the second experiment phase is discussed in terms of statistical power and a change in coping induced by the expectation of high intensity pain.

International audience; Head-mounted displays are now extensively developed and tested to be used in enhanced reality environments. The technique consists in transmitting synthetic visual information to the user's eyes in such a way that virtual objects are superimposed on the real world. Some displays give the feeling of viewing a rectangular head-fixed virtual screen with clearly visible contours. In addition, head-mounted displays allow for the presentation of head-fixed visual information to the user and this may be an advantage in some applications, military aeronautics for instance. Presenting head-fixed visual information may not be trivial with regard to actual models of spatial orientation. This study addressed the question of potential disorienting effects associated with head-mounted displays by investigating the influence of a head-fixed visual frame on the perception of the vertical when the head or the whole body was tilted in the frontal plane. In a first experiment, subjects were instructed to indicate the vertical by rotating a visual rod that appeared at the centre of the frame whilst tilting the head in various positions. This performance was compared with the effect on the subjective vertical of a tilted earth-based visual frame without head tilts as well as with the effect of tilting the head without a frame. With the tilted frames, subjects set the rod in an intermediate direction between the gravitational vertical and the orientation of the frame. Errors were substantially larger with a head-fixed visual frame during head tilt than with a tilted earth-based frame. This difference cannot be attributed to the addition of a postural effect caused by the head being tilted Moreover, continuous vision of the frame when its orientation changed improved performance only when the head and the frame were dissociated, i.e. with an earth-based frame. A second experiment investigated the effects of a head-fixed frame on the subjective vertical and on the voluntary control of head orientation when the whole body was tilted. The effect of a head-fixed frame was contrasted with the effect of a trunk-fixed frame. Results show that the head-fixed frame modified the head behaviour when subjects were instructed to align the head with the trunk. These errors contributed to an increase in the visual frame influence on the subjective vertical. Results of both experiments suggest that integrating visual information in the head-centric reference frame is crucial for spatial orientation. This Property of the perceptual system may be relevant for the design and use of head-mounted displays.

Spatial disorientation is a major source of accidents in aviation. Fighting such phenomenon necessarily entails the understanding of sensory systems that are involved in the perception of self-orientation, in order to adapt accordingly the cockpit displays. The introductory part of the manuscript sets out what is known about the occurrences of spatial disorientation. Some of the fundamental mechanisms of space perception are also presented, in relation to sensory ergonomics applied to the design of cockpit displays. The subsequent experimental studies deal at first with the influence of a head-fixed visual frame, integral to the head, on the perception of the vertical. The results mainly put forward the crucial role of the head-centric reference frame in the processing of oriented visual information. The potential implications of those results on the design of head-mounted displays, which are used in airplanes, are underlined. A second experimental chapter deals with the effects of galvanic vestibular stimulation on the perception of spatial orientation, a stimulation technique that could be included in the cockpit in order to fight spatial disorientation. It is demonstrated that such a stimulation can influence the subjective vertical, whatever the sensory modality involved. Besides, stimulating the vestibular apparatus gives rise to clearly dissociated effects on the subjective vertical, on the one hand, and the perception of body orientation, on the other hand. The results are compared with the current models of spatial orientation.; Les épisodes de désorientation spatiale forment une source importante d'accidents en aéronautique. Pour lutter contre ces phénomènes, il est nécessaire de comprendre les propriétés des systèmes sensoriels impliqués dans la perception de l'orientation spatiale, afin d'adapter au mieux l'instrumentation de bord. Le cadre général du présent manuscrit fait l'état des connaissances actuelles relatives à la désorientation spatiale. Il présente également certains des mécanismes fondamentaux de l'intégration multisensorielle dans le domaine de l'orientation spatiale, en relation avec l'ergonomie sensorielle des postes de pilotages. Les travaux expérimentaux qui suivent s'intéressent tout d'abord à l'influence d'un cadre visuel dont la caractéristique est d'être solidaire de l'orientation de la tête, sur la perception de la verticale. Les résultats suggèrent principalement le rôle fondamental du référentiel céphalocentré dans le traitement des informations visuelles orientées. Les répercutions potentielles de ces résultats sur la conception des afficheurs de casque, utilisés dans les aéronefs, sont envisagées. Une seconde partie expérimentale s'intéresse aux effets de la stimulation galvanique vestibulaire sur la perception de l'orientation spatiale, avec la perspective d'inclure ce type de stimulation dans les postes de pilotage. Il est montré que cette technique de stimulation influence la verticale subjective, quelle que soit la modalité sensorielle impliquée. De plus, les stimulations de l'appareil vestibulaire induisent des effets clairement dissociés sur la verticale subjective, d'une part, et sur la perception de l'orientation du corps, d'autre part. Ces résultats sont mis en rapport avec les modèles actuels de l'orientation spatiale.

International audience

Remote collaborative situations in industry involve new constraints for workers. In the context of using immersive virtual environments to collaborate, we set up a series of experiments focusing on the mutual comprehension, and more specifically on the process of sharing spatial information. Results of the first experiment showed the influence of one collaborator’s role on spatial statements. Guides and manipulators both used statements preferentially centered on manipulator’s action, which supports the least collaborative effort principle. Results of two experiments about spatial statements production allowed to point out that mental workload is modulated by the target position relative to the addressee. According to the target location, the speaker must operate or not cognitively costly mental transformations to take the addressee perspective. However this workload could be lowered by means of visual cues. The last experiment focused on the understanding of spatial statements. It showed that statements centered on the addressee are the easiest to understand when the task is also centered on the addressee. Some exocentered statements could also induce a lower mental workload but only in some conditions. Results are discussed in relation to the least collaborative effort principle and the spatial frames of reference theory. This work opens new leads to facilitate remote collaboration through virtual environments.; Les situations de collaboration à distance dans l’industrie induisent de nouvelles contraintes pour les opérateurs. Dans le contexte de l’utilisation d’environnements virtuels collaboratifs immersifs, nous avons mis en place une série d’études portant sur la compréhension mutuelle, et plus particulièrement au partage d’information de nature spatiale. Les résultats de la première étude ont permis de mettre en évidence l’influence du rôle des participants, guides ou manipulateurs, sur les énoncés spatialisés. Les énoncés étaient centrés préférentiellement sur l’action du manipulateur, suggérant la recherche d’un moindre effort collaboratif. Les résultats de deux études sur la production d’énoncés spatialisés ont permis d’identifier que l’exigence mentale pour la production d’énoncés est modulée notamment par la position de la cible à décrire par rapport au destinataire. En effet, selon la position de la cible, le locuteur doit opérer ou non des transformations mentales coûteuses pour prendre la perspective du destinataire. Cet effort peut être amoindri en présence d’indices visuels distaux. La dernière étude, portant sur la compréhension d’énoncés spatialisés, a permis de mettre en évidence que les énoncés centrés sur le destinataire sont les plus simples à comprendre pour une tâche de nature égocentrée. Certains énoncés exocentrés induisent également une moindre exigence mentale, mais uniquement selon certaines conditions. Les résultats sont discutés selon le principe du moindre effort collaboratif et la théorie des cadres de référence. Ce travail a permis de proposer des pistes de développement pour faciliter la collaboration à distance dans les environnements virtuels.

This work concerned the assessment of drivers’ attentional state in highly automated vehicle. A specific attention had been placed on the analysis of driver’s visual strategies. The first two experiments dealt on the link between distraction, visual strategies and takeover quality. To that end, divers had to realize non-driving activities during automated driving and had to deal with a non-planned takeover in obstacle avoidance situation. Results showed that there was an effect of the gaze location at the time of the takeover request but also an effect of the development of situational awareness on the takeover quality. The last experiment looked at the effect of different automation modes on driver inattention in the form of mindwandering. The evolution of mindwandering over time has been estimated by modeling the driver’s visual strategies. Results showed that mindwandering increased strongly to reach high levels very quickly and then fluctuated around a stable trend. These results are discussed in the context of the development of driver monitoring systems for autonomous vehicle.; Ce travail a porté sur l'évaluation de l'état attentionnel de conducteurs à bord d'un véhicule hautement automatisé. Un accent particulier a été mis sur l'analyse des stratégies visuelles du conducteur. Les deux premières expérimentations ont porté sur le lien entre distraction, stratégies visuelles et qualité de la reprise en main. Pour ce faire, les conducteurs ont réalisé des activités non liées à la conduite pendant une phase de conduite autonome et ont dû faire face à une demande de reprise en main non planifiée dans une situation d'évitement d'obstacle. Les résultats montrent qu'il existe un effet de la localisation du regard au moment de la demande de reprise en main mais également du niveau de l’élaboration de la conscience de la situation sur la qualité de la reprise en main. La dernière expérimentation s'est intéressée à l'effet de différentes modalités d'automatisation sur l'inattention du conducteur, prenant la forme d'un vagabondage de la pensée. L'évolution du vagabondage au fil du temps a été estimé en modélisant les stratégies visuelles des conducteurs. Les résultats montrent que le vagabondage augmente très fortement pour atteindre des niveaux élevés très rapidement puis fluctue autour d'une tendance stable. Ces résultats sont discutés dans la perspective du développement des outils d'évaluation de l'état du conducteur à bord des véhicules autonomes.

Risk perception is a crucial aspect that needs to be investigated in order to promote the acceptance of autonomous vehicles in shared spaces with pedestrians. This thesis is structured around three empirical contributions. The first contribution focuses on the possibility of modelling the dynamics of subjective risk perception using an impulse response. The performance obtained is rather uneven and reflects a large inter-individual variability in the risk assessment strategy. The other two contributions demonstrate that risk perception can be measured in real time and in complex dynamic interactions using both subjective and electrodermal response measures. The results of these contributions are based on Bayesian network modelling that allows for the simultaneous analysis of relationships between dynamic factors and risk perception measures. The conclusions of this work provide theoretical elements on the links between the implicit and explicit dimensions of risk perception, as well as concrete elements to take into account human factors in the development of navigation behaviors of autonomous vehicles.; La perception du risque est un aspect crucial qu'il est nécessaire d'étudier dans le but de favoriser l'acceptation des véhicules autonomes dans des espaces partagés avec des piétons. Ce travail s'articule autour de trois contributions empiriques. La première contribution porte sur la possibilité de modéliser la dynamique de perception subjective du risque par le biais d'une réponse impulsionnelle Les performances obtenues sont assez inégales et reflètent une grande variabilité interindividuelle dans la stratégie d'évaluation du risque. Les deux autres contributions mettent en évidence que la perception du risque peut se mesurer en temps réel et dans des interactions dynamiques complexes en s'appuyant à la fois sur des mesures subjectives et des mesures de réactions électrodermales.Les résultats de ces contributions sont basés sur une modélisation par réseau bayésien permettant d'analyser simultanément les relations entre facteurs dynamiques et mesures de perception du risque. Les conclusions de ces travaux apportent des éléments théoriques sur les liens entre les dimensions implicite et explicite du risque ressenti, ainsi que des éléments concrets permettant de prendre en compte le facteur humain dans le développement des comportements de navigation des véhicules autonomes.

Driver behaviour, including visual strategies, can be modified in autonomous driving situations compared to manual driving.The first axis of this PhD work focuses on the adaptation of visual strategies in autonomous driving. To this end, two experiments were conducted in different road contexts (rural roads with bends and motorway), with drivers having to continuously or partially supervise the driving scene.The results show that the strategies evolve in all cases towards a disengagement of the gaze from the immediate road to move either upstream for more anticipation (continuous supervision on rural roads) or towards irrelevant areas (partial supervision on motorways). The second line of work deals with the link between visual strategies and the out-of-the-loop phenomenon, i.e. the gradual disenga-gement of the driver from supervision as he drives in autonomous mode. The results show that it is possible to predict the state of the driver in the loop from his spontaneous visual strategies.These conclusions are based on statistical modelling work. The resulting models highlight the importance of considering indicators of gaze dynamics to predict the driver's state.These conclusions are based on statistical modelling work. The resulting models highlight the importance of considering indicators of gaze dynamics to predict the driver's state. These results can be used to develop algorithms for detecting driver disengagement in future autonomous vehicles.

Driver behaviour, including visual strategies, can be modified in autonomous driving situations compared to manual driving.The first axis of this PhD work focuses on the adaptation of visual strategies in autonomous driving. To this end, two experiments were conducted in different road contexts (rural roads with bends and motorway), with drivers having to continuously or partially supervise the driving scene.The results show that the strategies evolve in all cases towards a disengagement of the gaze from the immediate road to move either upstream for more anticipation (continuous supervision on rural roads) or towards irrelevant areas (partial supervision on motorways). The second line of work deals with the link between visual strategies and the out-of-the-loop phenomenon, i.e. the gradual disenga-gement of the driver from supervision as he drives in autonomous mode. The results show that it is possible to predict the state of the driver in the loop from his spontaneous visual strategies.These conclusions are based on statistical modelling work. The resulting models highlight the importance of considering indicators of gaze dynamics to predict the driver's state.These conclusions are based on statistical modelling work. The resulting models highlight the importance of considering indicators of gaze dynamics to predict the driver's state. These results can be used to develop algorithms for detecting driver disengagement in future autonomous vehicles.

The evolution of technology and the shift in the role devoted to operators in industry, from simple task execution to complex system control, provoke a development of the literature concerning assistance to fabrication process. This thesis of psychology joins this literature. It aims at designing an assistant to fabrication process, called the virtual companion, which would both support performances and maintain or improve competences. In particular, the virtual companion would be addressed to operators in an aeronautical factory. The case study requires in line simulation of dynamic physical processes. Two fields of the literature were used: frameworks concerning interface design for dynamic situations and those concerning competences. A three-parts methodology was used: (1) activity analysis of the operators in the real factory, (2) conception and validation of a microworld, and (3) design and test of a virtual companion in the microworld. Results show that the operators of the factory use a misleading scheme during their activity. This misleading scheme is also used by random participants. Moreover, this misleading scheme has an effect on the performances of random participants trained in the microworld. Finally, the virtual companion, whose interface is design based on EID principles, does not eliminate the misleading scheme. However, the participants who did not use the misleading scheme performed better after using the virtual companion. This results leads to the conclusion that studying the operators’ use of misleading schemes seems accurate as such schemes may impair the use of interfaces designed to support their activity.; L’évolution des technologies et la mutation du rôle attribué aux opérateurs de l’industrie suscitent un essor de la littérature concernant l’assistance aux procédés de fabrication. Cette thèse de psychologie s’inscrit dans cette littérature. Son objectif est de concevoir un assistant de contrôle aux procédés de fabrication, le compagnon virtuel, qui permette de maintenir à la fois les performances et les compétences des opérateurs. Le système étudié, une usine aéronautique, nécessite le recours à la simulation de processus physiques dynamiques. Deux champs de la littérature sont mobilisés : les cadres théoriques autour de la conception d’interfaces pour les environnements dynamiques et ceux sur le développement des compétences. Une méthodologie en trois parties est utilisée : (1) analyse de l’activité des opérateurs dans l’atelier réel, (2) validation d’un micromonde et (3) test d’un compagnon virtuel dans le micromonde. Les résultats montrent que les opérateurs de l’atelier réel mobilisent un schème trompeur lors de leur activité. Ce schème trompeur est également présent chez des participants tout-venants et il a un effet sur les performances des participants entrainés dans le micromonde. Enfin, le compagnon virtuel, conçu selon les principes de l’EID, ne permet pas de faire disparaitre le schème trompeur. Cependant, il permet aux participants qui n’utilisent pas le schème trompeur d’obtenir de bonnes performances. Les résultats obtenus étayent l’idée qu’il est pertinent de vérifier si des opérateurs ne mobilisent pas des schèmes trompeurs qui pourraient être un frein à l’utilisation des interfaces proposées pour les assister.

Driving simulation, ESC, Sensory evaluation, Subjective evaluation, Steering control; Evolutions of the design process tend to the increasing integration of digital tools, including driving simulators, in the validation loop. It is therefore necessary to build methods of uses fitted to meet these new constraints and explore new possibilities that these tools can offer. The evaluation of the driving dynamics performances is discussed here in the case of trajectory control in turn subjected to disturbances. The objective of this work is to propose a methodology for subjective evaluation, on simulator, of performances related to vehicle dynamics and apply it to the characterization of a trajectory correction system. Experimental studies on simulator, in the presence and in the absence of inertial restitution, were conducted to validate a methodology for assessment and identify sensory cues put in. A model based on classical psychophysical relations has been proposed to link the subjective judgments and the objective data of the dynamic behavior of the vehicle. Finally, the use of this model has been used to characterize the onset of a stability control system (ESC). This work has been carried out in the framework of a CIFRE PhD (ANRT contract No 1181/2008) at RENAULT, Technical Center for Simulation and at Research Institute of Communication and Cybernetics of Nantes (IRCCyN), UMR 6597, under the co-direction of MM. Jean-François Petiot (IRCCyN - Ecole Centrale de Nantes) and Andras Kemeny (RENAULT) and supervision of Mr. Frank Mars (IRCCyN - CNRS).; Les évolutions des processus de conception tendent à l'intégration grandissante des outils numériques, notamment les simulateurs de conduite, dans les boucles de validation. Il est donc nécessaire de construire des méthodologies d'usages adaptées pour répondre à ces nouvelles contraintes et d'explorer les nouvelles possibilités que ces outils peuvent offrir. L'évaluation des prestations dynamiques de conduite est abordée ici dans le cas du contrôle de la trajectoire en virage soumis à des perturbations. L'objectif de ces travaux est de proposer une méthodologie d'évaluation subjective, sur simulateur, des prestations liées à la dynamique du véhicule et de l'appliquer pour la caractérisation d'un système de correction de trajectoire. Des études expérimentales sur simulateur, en présence et en l'absence de restitution inertielle du mouvement, ont été conduites pour valider une méthodologie d'évaluation et identifier les indices sensoriels mis en jeu. Une modélisation basée sur les relations psychophysiques classiques a été proposée pour relier les jugements subjectifs aux données objectives du comportement dynamique du véhicule. Enfin, l'utilisation de cette modélisation a été utilisée pour caractériser le déclenchement d'un système d'aide au contrôle de la trajectoire (ESC). Ces travaux ont été réalisés dans le cadre d'une thèse CIFRE (contrat ANRT n° 1181/2008) à RENAULT, Centre Technique de Simulation et à l'Institut de Recherche en Communication et Cybernétique de Nantes (IRCCyN), UMR 6597, sous la co-direction de MM. Jean-François Petiot (IRCCyN - Ecole Centrale de Nantes) et Andras Kemeny (RENAULT) et l'encadrement de M. Franck Mars (IRCCyN - CNRS).

The work in this thesis employs the Human-Machine Cooperation framework to study the relationship between symbolic (interpretive) processing of the context and subsymbolic processing of the intervention of driving assistance devices that act directly on the steering wheel. Four experimental studies on a driving simulator are presented.The first two focus on the ability of drivers to remain in full control of the vehicle when the assistance occasionally intervenes on the steering wheel, in case of imminent risk of lane departure (haptic priming). The results show that the driver’s reaction times are fully determined at the sensorimotor level. However, symbolic processes are involved as well at a very early stage to inhibit or modulate the motor response. Two other studies focus on Human-Machine Cooperation when the intervention of the device is continuous (shared control). The optimal level of shared control and the medium-term adaptation of the driver to the automation are studied. The results show that it is possible to effectively integrate the intervention of automation into the drivers sensorimotor control loops. However, successful integration requires respecting certain constraints, such as the fine calibration of the level of automation and the timing of the intervention; Ce travail de thèse utilise le cadre de la Coopération Homme-Machine afin d’étudier l’articulation entre le traitement symbolique (interprétatif) du contexte et le traitement subsymbolique de l’intervention d’assistances à la conduite directement sur le volant. Quatre études expérimentales menées sur simulateur de conduite sont présentées.Les deux premières études s’intéressent à la capacité des conducteurs à conserver la totale maîtrise du véhicule lorsque l’assistance intervient ponctuellement sur le volant, en cas de risque imminent de sortie de voie (amorçage haptique). Les résultats montrent que seuls les temps de réaction des conducteurs sont entièrement déterminés au niveau sensori-moteur, les processus symboliques intervenant très tôt pour inhiber ou moduler la réponse motrice. Les deux dernières études s’intéressent à la Coopération Homme-Machine lorsque l’intervention de l’automate sur le volant est continue (contrôle partagé). Le degré de partage entre les deux agents et l’adaptation à l’assistance sur le moyen terme ont été étudiés. Les résultats permettent de conclure qu’il est possible d’intégrer efficacement l’intervention des automatesdans les boucles de contrôle sensori-motrices à condition que cette intégration respecte certaines précautions. Il s’agira notamment de calibrer finement le degré d’intervention ou la temporalité d’intervention dans ces boucles

In order to facilitate driving and prevent accidents due to lane departure, this thesis focuses on the development of an assistance device for lane keeping of a passenger vehicle. The review of the literature on this subject has highlighted the need for a better understanding of the interactions between the driver and the vehicle-road system. The goal is to develop a shared control mode based on the anticipation of risk and the prediction of the most probable actions of the driver. The thesis was mainly articulated around two axes: the cybernetic modeling of the driver in his task of lateral control of the vehicle, and the design of a shared steering control using a Driver-Vehicle-Road model (DVR). The proposed driver model is consistent with what is known about sensorimotor and cognitive control in humans. It represents the anticipatory and compensatory visual control of steering, as well as the execution of steering actions by the neuromuscular system. It has been identified from experimental data collected on a fixed-base driving Simulator SCANeRTM. The model showed a good fit to the driver behavior, supporting later the development of the global model DVR. Concerning the shared control, two experimental studies were performed. The first (socalled CoLat1) is similar to already existing LKS systems. CoLat1 is primarily designed as an "electronic pilot" that has to drive alone the vehicle. Without using a driver model, it results from applying the H2-preview control synthesis to a vehicle-road model. Shared steering is performed only in the implementation phase: the control that it provides is partially applied to the steering wheel leaving the complementary control part to be applied by the driver. The second (so-called CoLat2) addresses a shared control resulting from applying the H2-preview control synthesis to the global model (DVR). The H2/LQ-Preview control law has been used in the both cases (CoLat1 & 2) because it allows to include knowledge of the future road curvature and to consider relevant performance criterions. A generalization of the H2-Preview solution was accomplished in the case where the disturbance signal is predicted beyond the preview horizon on the basis of a generator model. Innovative criterions were proposed for assessing the risk of lane departure, the sharing level between the driver and the lateral assistance, as well as its cooperative or conflicting behavior. They have been formalized through the criteria supporting the synthesis of CoLat2, enabling therefore a precise management of trade-off between the quality of lane keeping and the quality of driver-assistance interaction. CoLat2 has shown better performance compared to CoLat1 in terms of driver-assistance cooperation thanks to "online" feedback of the estimated driver state. The μ-Analysis of the overall system robustness with respect to the driver model uncertainties has shown that the stability is guaranteed for a wide range of parametric variations.; Afin de faciliter la conduite et prévenir les accidents par sortie de voie, ce travail porte sur le développement d'un dispositif d'assistance au contrôle latéral de la trajectoire d'un véhicule automobile. L'état de l'art des travaux portant sur ce thème a mis en évidence la nécessité de mieux prendre en compte l'interaction du conducteur avec l'ensemble véhicule-route. Ceci nous a conduit à développer un mode de contrôle partagé basé sur l'anticipation du risque et sur la prédiction des actions les plus probables du conducteur. Le travail de la thèse a été articulé principalement autour de deux axes : la modélisation cybernétique du conducteur dans sa tâche de contrôle latéral, et la conception du contrôle partagé de la direction sur la base du modèle global CVR (conducteur-véhicule-route). Un modèle du conducteur a été développé en cohérence avec les connaissances actuelles sur la sensorimotricité humaine. Ses paramètres ont été identifiés en utilisant des données expérimentales issues d'un simulateur de conduite SCANeRTM. Ensuite, Deux stratégies du contrôle partagé ont été conçues et évaluées expérimentalement. La première (nommée CoLat1) se repose sur un principe similaire à celui des assistances LKS commercialisées à ce jour. Sa conception repose sur l'utilisation d'un modèle véhicule-route, et ne fait appel à aucun modèle du conducteur. Le partage n'est fait que de la mise en œuvre en appliquant partiellement la commande qu'il produit. La deuxième (nommée CoLat2) repose sur une conception tenant compte des modalités du partage de la conduite avec le conducteur. Sa conception repose sur l'utilisation d'un modèle global (CVR). La commande H2/LQ anticipative (with preview) a été choisie dans les deux cas (CoLat1 & 2) pour ce qu'elle permet la prise en compte de la connaissance anticipée de la courbure de la route et la définition de critères de performance pertinents. Elle a été revisitée pour tenir compte de modèles prédicteur des signaux perturbateurs.

In order to facilitate driving and prevent accidents due to lane departure, this thesis focuses on the development of an assistance device for lane keeping of a passenger vehicle. The review of the literature on this subject has highlighted the need for a better understanding of the interactions between the driver and the vehicle-road system. The goal is to develop a shared control mode based on the anticipation of risk and the prediction of the most probable actions of the driver. The thesis was mainly articulated around two axes: the cybernetic modeling of the driver in his task of lateral control of the vehicle, and the design of a shared steering control using a Driver-Vehicle-Road model (DVR). The proposed driver model is consistent with what is known about sensorimotor and cognitive control in humans. It represents the anticipatory and compensatory visual control of steering, as well as the execution of steering actions by the neuromuscular system. It has been identified from experimental data collected on a fixed-base driving Simulator SCANeRTM. The model showed a good fit to the driver behavior, supporting later the development of the global model DVR. Concerning the shared control, two experimental studies were performed. The first (socalled CoLat1) is similar to already existing LKS systems. CoLat1 is primarily designed as an "electronic pilot" that has to drive alone the vehicle. Without using a driver model, it results from applying the H2-preview control synthesis to a vehicle-road model. Shared steering is performed only in the implementation phase: the control that it provides is partially applied to the steering wheel leaving the complementary control part to be applied by the driver. The second (so-called CoLat2) addresses a shared control resulting from applying the H2-preview control synthesis to the global model (DVR). The H2/LQ-Preview control law has been used in the both cases (CoLat1 & 2) because it allows to include knowledge of the future road curvature and to consider relevant performance criterions. A generalization of the H2-Preview solution was accomplished in the case where the disturbance signal is predicted beyond the preview horizon on the basis of a generator model. Innovative criterions were proposed for assessing the risk of lane departure, the sharing level between the driver and the lateral assistance, as well as its cooperative or conflicting behavior. They have been formalized through the criteria supporting the synthesis of CoLat2, enabling therefore a precise management of trade-off between the quality of lane keeping and the quality of driver-assistance interaction. CoLat2 has shown better performance compared to CoLat1 in terms of driver-assistance cooperation thanks to "online" feedback of the estimated driver state. The μ-Analysis of the overall system robustness with respect to the driver model uncertainties has shown that the stability is guaranteed for a wide range of parametric variations.; Afin de faciliter la conduite et prévenir les accidents par sortie de voie, ce travail porte sur le développement d'un dispositif d'assistance au contrôle latéral de la trajectoire d'un véhicule automobile. L'état de l'art des travaux portant sur ce thème a mis en évidence la nécessité de mieux prendre en compte l'interaction du conducteur avec l'ensemble véhicule-route. Ceci nous a conduit à développer un mode de contrôle partagé basé sur l'anticipation du risque et sur la prédiction des actions les plus probables du conducteur. Le travail de la thèse a été articulé principalement autour de deux axes : la modélisation cybernétique du conducteur dans sa tâche de contrôle latéral, et la conception du contrôle partagé de la direction sur la base du modèle global CVR (conducteur-véhicule-route). Un modèle du conducteur a été développé en cohérence avec les connaissances actuelles sur la sensorimotricité humaine. Ses paramètres ont été identifiés en utilisant des données expérimentales issues d'un simulateur de conduite SCANeRTM. Ensuite, Deux stratégies du contrôle partagé ont été conçues et évaluées expérimentalement. La première (nommée CoLat1) se repose sur un principe similaire à celui des assistances LKS commercialisées à ce jour. Sa conception repose sur l'utilisation d'un modèle véhicule-route, et ne fait appel à aucun modèle du conducteur. Le partage n'est fait que de la mise en œuvre en appliquant partiellement la commande qu'il produit. La deuxième (nommée CoLat2) repose sur une conception tenant compte des modalités du partage de la conduite avec le conducteur. Sa conception repose sur l'utilisation d'un modèle global (CVR). La commande H2/LQ anticipative (with preview) a été choisie dans les deux cas (CoLat1 & 2) pour ce qu'elle permet la prise en compte de la connaissance anticipée de la courbure de la route et la définition de critères de performance pertinents. Elle a été revisitée pour tenir compte de modèles prédicteur des signaux perturbateurs.