Μελέτη και μοντελοποίηση αλληλεπιδράσεων οδηγών με βάση τη θεωρία λεκτικών πράξεων, προς την κατεύθυνση εμπλουτισμού συστημάτων αυτοματοποίησης της οδήγησης

Πορτούλη, Ευαγγελία

dc.contributor.author	Πορτούλη, Ευαγγελία	el
dc.date.accessioned	2015-09-08T07:37:01Z
dc.date.available	2015-09-08T07:37:01Z
dc.date.issued	2015-09-08
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/41203
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.2063
dc.rights	Default License
dc.subject	Επικοινωνία, Αλληλεπιδράσεις οδηγών, Αυτοματοποίηση οδήγησης, Γλωσσολογικό μοντέλο, Πρόβλεψη πρόθεσης	el
dc.title	Μελέτη και μοντελοποίηση αλληλεπιδράσεων οδηγών με βάση τη θεωρία λεκτικών πράξεων, προς την κατεύθυνση εμπλουτισμού συστημάτων αυτοματοποίησης της οδήγησης	el
dc.contributor.department	Βιομηχανικής Διοίκησης και Επιχειρησιακής Έρευνας	el
heal.type	doctoralThesis
heal.classification	Εργονομία	el
heal.language	el
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2015-07-21
heal.abstract	This thesis studied the drivers’ communicative interactions, which constitute a necessary component of the driving task. These interactions are based on successive elementary communicative acts. It is a common observation that when drivers predict that their vehicle may be found in the future in dangerous proximity to another vehicle, they tend to intentionally communicate in advance their motion intent to the other involved drivers, so that they may jointly plan and execute a safe future motion plan. At a theoretical level, the objective of this thesis was to study the interactive part of the driving task, shedding light on the phenomenon of communication of drivers’ intent and their agreement on a future common motion plan. At a practical level, the aim of the thesis was to develop a model of drivers’ communicative interactions, for integration, as a manoeuvring negotiations level, in Driving Support and Automation Systems (DSAS), so that their functionality is more in accordance to human expectations. Chapter 2 studies the importance of the studied phenomenon of communicative drivers’ interactions, and how such interactions can be decomposed into elementary phases in the form of a flowchart and described in the form of a discussion between the involved drivers. Video recordings of natural overtakings on a highway, with two directions without central barrier, were used for a first empirical confirmation of the frequency and significance of such interactions. The analysis of 82 overtakings showed that the intent to overtake could be anticipated before the start of the manoeuvre in 45,12% of cases, through formal and informal cues. In 6 of the 22 cases when the intent could be anticipated and there was an oncoming vehicle, its driver modified the trajectory of the vehicle so as to facilitate the overtaking. Chapter 3 includes a literature review of cognitive drive models, such as the hierarchical model of Michon, the model of Wilde, the COCOM model, the situation awareness levels of Endsley, the ECOM model and the Joint Action Control model. All but one of these models are personal models, that focus on a single driver, and analyse how he/she performs the various activities necessary for the driving task. There is no clear reference or modelling of deliberate drivers’ communicative interactions before a manoeuvre, which aim to ensure their common understanding for the safer and more convenient manoeuvre conduct. Chapter 4 presents the evolution of attempts to design and develop DSAS and systems to recognise driver’s intent. The initial design attempts were based on parameters such as time to collision with the involved vehicle. Still, experience showed that warnings by such DSAS were not in agreement with drivers’ expectations. Thus, the recent design guidelines by the National Highway Traffic Safety Administration suggest that the timing of the warnings by DSAS should be based on the expected driver’s response time and on the deceleration level selected by the driver. So, several DSAS estimate risk of collision based on the forecasted future vehicles trajectory according to motion models. But, the warning distances calculated by such DSAS are large and not consistent with the risk assessment by an average driver. Indeed, several studies report that the acceptance of warning systems by drivers is rather limited. One possible reason is that the warnings of DSAS are not consistent with the drivers’ estimations and expectations. In this thesis it is argued that this occurs because DSAS do not consider in their design the drivers’ communicative interactions. Chapter 5 attempts to clarify some terms used when describing the phenomenon. Coordination can be regarded as the orchestration of behaviours so as to achieve a common aim and is considered to occur at a pre-conscious level. Cooperation can be regarded as an observed social interaction with a positive outcome for each of the involved players. Social behaviour is the behaviour which is oriented towards other people and social interactions are the acts, actions or practices of two or more “selves”, each oriented towards one another, i.e. any behaviour which seeks to influence or take into account the subjective experiences or intent of another self. In this thesis it is argued that driving is often a social activity, that drivers consider other vehicles as social units, as "animate human-vehicles", and largely base their expectations for the evolution of the traffic scene by observing and interpreting the behaviour of these social units. Then, it is analysed how driving can be considered as a joint activity, based on a Basic Agreement, which for driving can be considered to be the safe and smooth traffic on the road network for all road users. A joint activity also requires the existence of Common Ground, which for driving is based on the explicit rules of the Traffic Law but also on commonly accepted conventions, as argued in this thesis. It is also supported that when there is a risk for a breakdown of the Common Ground, when drivers are uncertain about others’ intent, they consciously seek to update and repair their Common Ground, trying to interact with other drivers, so as to communicate their motion intent and to agree on a safe future motion plan. Chapter 6 analyses how a behaviour which the observer can describe in semantic terms can be considered as a linguistic behaviour. Then, the Speech Act Theory of Austin (1962) and the classification of illocutionary acts by Searle (1975) are presented, which can be used for modelling non-verbal communicative acts, such as those studied in this thesis. In this framework, drivers’ communicative interactions can be modelled using the locutionary / illocutionary / perlocutionary typology by Austin. It is argued in this thesis that any communicative act in such interactions includes three types of acts, a "locutionary", an illocutionary and a perlocutionary one. The "locutionary" act conveys the sense and reference according to Austin. Such acts may be the use of the horn, the flashing of headlights, the lateral displacement of the vehicle towards the central lane marking or the deliberate close following of the lead vehicle. The illocutionary act conveys a certain force, for example “request” to empty the lane or “frightening” the lead “human-vehicle” so as to persuade him/her to change lane. The perlocutionary act is the act done due to the communicative act, for example the lead “human-vehicle” facilitates the overtaking or the lead “human-vehicle” empties the lane. A sequence of illocutionary acts in drivers’ communicative interactions is in this thesis modelled as a "Conversation" using the Linguistic Model of Drivers’ communicative Interactions (LMDI). In such an interaction, drivers use and interpret certain communicative acts or cues at the illocutionary level as "Directives", for example "requests" or "commands", and other acts or cues as "Commissives", for example "acceptances" or "rejections" of the "Directives". Regarding the non-verbal "locutionary" acts, the communication of intent between drivers can be mediated by different, formal and informal, communicative acts or cues. It is argued in this thesis that drivers collect information to anticipate the intent of other "human-vehicles" from two levels: a) the social level, which includes the explicit, intentional communicative acts emitted by a driver, and b) the physical level, which includes communicative acts emitted intentionally or cues used to interpret intent based on the observed motion of a "human-vehicle" on the road. Chapter 7 presents a first empirical verification of the LMDI based on observations of natural drivers’ interactions in real traffic conditions. The observations took place in the Hymettus ring road, a road with at least two lanes per direction and a central barrier. The drivers were asked to drive normally their own vehicle while in parallel commenting aloud about their observations in the surrounding traffic and any action undertaken by themselves as a result of any such observation. During the route one observer was seated in the passenger seat and recorded with two cameras the traffic scene in front and behind the vehicle and the driver’s commentary. The observations of 25 experienced drivers, 17 men and 8 women, were analysed. The findings support the argument that in many cases, drivers consider other vehicles as social units, as "animate human-vehicles", and largely base their expectations for the evolution of the traffic scene by observing and interpreting the behaviour of these social units. Indeed, on average in 38.7% of the uttered periods drivers referred to observations and interpretations of social behaviour of "animate human-vehicles". From the 360 periods referring to observations of social behaviour, 90 referred to anticipation of others’ intent. According to the analysis, in order to predict the future trajectory of a moving vehicle, a driver uses the physical geometry of the road and the physical limitations, assuming a smooth motion of the vehicles in the future. Any observed deviation of a vehicle from its expected smooth motion, is considered as a possible result of an intentional action by its particular driver and is used to anticipate intent of this “human-vehicle". These cues, which are created by observing inanimate objects, are supplemented by the cues emitted by the animate human drivers, and contribute to the anticipation of intent. In addition, the observer anticipates intent based on stereotypes. Furthermore, according to the analysis, the illocutionary force of acts within the observed interactions could be modelled as a "discussion" with a series of illocutionary acts, as represented by the LMDI. Empirical findings also support the argument that in cases of uncertainty, when the underlying Common Ground is not enough for safe prediction of the evolution of the traffic scene, drivers consciously seek to interact with other drivers, so as to communicate their intent, to update and repair their Common Ground and to agree on a safe future motion plan. Indeed, the analysis showed that most of the communicative interactions were initiated when one driver wished to change lane in the presence of another “human-vehicle” in the target lane or when one driver wished to drive faster than a lead “human-vehicle”. Finally, the observations were analysed separately for older and younger drivers. According to the analysis, older drivers face driving as a social phenomenon at the same level as younger drivers. However, the communicative acts and communicative interactions clearly tended to be less for older drivers. These findings support the argument that older drivers may adopt a strategy that allows them to avoid frequent communicative interactions, in order to compensate for their degraded abilities. Chapter 8 presents a first assessment of the effects from the integration of a manoeuvring negotiations unit based on the LMDI in DSAS. The assessment was based on an experiment using a dynamic driving simulator. Four different driving scenarios were algorithmically developed and four driving support systems were simulated, a system warning about collision risk during lane change on highways, a system warning about crash risk with the oncoming car in case of left turn on bi-directional urban road, a system warning about collision risk while entering a highway, and a system warning about crash risk with the oncoming car in case of overtaking on a rural road. For each system, two conditions were simulated. In the first condition, the warning was provided according to the calculated Time To Collision with the other vehicle involved in the manoeuvre, at two levels, high and medium risk. In the second condition, a manoeuvring negotiations unit, based on the LMDI, was additionally simulated. More precisely, the software simulated the “acceptance” by the other “driver” of the participant’s manoeuvring “request”, which was supposedly signalled by the activation of the turn indicator by the participant. The "acceptance" was signalled by an additional icon besides the collision risk icons which were displayed according to the Time To Collision value. The study revealed some effects on parameters relevant to traffic efficiency due the possibility of such negotiations. More specifically, in three of the four scenarios the participants started their manoeuvre earlier and at conditions of greater objective risk when there was the "explicit" consent of the other “driver” to their intended manoeuvre. This suggests that participants felt more confident when the system was providing the explicit "consent" of the other "driver" than when the warning was provided only in accordance to the laws of physics (i.e. the Time To Collision value). The found effects on driving behaviour and the subjective assessments of drivers as regards the functionality of the systems in both conditions support the argument that drivers feel more certain when there is the possibility of communicating with other drivers, i.e. when there is an explicit "consent" by others to their intended manoeuvre. The findings support the argument that the functionality of DSAS after the integration of a manoeuvring negotiations level is more in accordance with human assessments and expectations. Finally, Chapter 9 summarizes the general conclusions of this thesis, and specifies directions for further research. The results of this thesis show that: (i) social interactions are and will remain an important component of the driving task and (ii) DSAS can be enriched by the integration of a social interactions level based on a model such as the LMDI. Such an endeavour, if successful, will ensure that the performance and functionalities of DSAS are more in accordance to that of the human driver and the human driver’s expectations. In the future, it will be necessary to collect further empirical data from communicative interactions through natural observations of more manoeuvre types in different traffic environments, so as to validate or expand the issues identified. Furthermore, future studies of such negotiations modules should involve two or more interacting human drivers and should record and analyse full cycles of communicative interactions in a more naturalistic setting, while experiments including automated vehicles should be also conducted. Future studies should also focus more specifically on the design of the manoeuvring negotiations module, i.e. how to handle cases when there will be no “response” to the manoeuvring “request” or when the observed behaviour of the vehicle is not in line with the expected one according to the “response” given to the manoeuvring “request”, so as to protect the whole system from any malicious or irresponsible use.	en
heal.advisorName	Μαρμαράς, Νίκος	el
heal.committeeMemberName	Ναθαναήλ, Δημήτρης	el
heal.committeeMemberName	Λεώπουλος, Βρασίδας	el
heal.committeeMemberName	Παπακωστόπουλος, Βασίλης	el
heal.committeeMemberName	Γκόλιας, Ιωάννης	el
heal.committeeMemberName	Γιαννόπουλος, Γεώργιος	el
heal.committeeMemberName	Κοντογιάννης, Θωμάς	el
heal.academicPublisher	Σχολή Μηχανολόγων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	197
heal.fullTextAvailability	true