Juan M. Belda Lois *, Sofía Iranzo Egea, Nicolás Palomares Olivares, José Laparra Hernández, José S. Solaz Sanahuja, Elisa Signes i Pérez

Instituto de Biomecánica de Valencia (IBV) Universitat Politècnica de València (Edificio 9C) Camino de Vera s/n (E-46022) Valencia (Spain).

* IBV’s Healhcare Technology Group. CIBER in Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN)

The acceptability of connected and autonomous vehicles is a complex process involving several factors including individual, social, cognitive and perceptive factors, among others. The SUaaVE project has built a model for evaluating emotions in order to integrate the user's emotions and sensations into the behavior of an autonomous vehicle. The basis of this model is a set of rules that determine the emotion that a person is feeling, based on environmental data and data from the individual themself.

INTRODUCTION

The acceptability of connected and autonomous vehicles is a complex process involving several factors including individual, social, cognitive and perceptive factors, among others. Aspects such as confidence in the autonomous vehicle, safety, control, satisfaction and convenience seem to play a crucial role in defining the acceptability of this new vehicle concept. 

As a result, not only must autonomous vehicles meet technical performance specifications, but they must also satisfy social expectations in terms of behavior towards other road users: drivers of conventional vehicles, passengers or pedestrians.

In accordance with this hypothesis, the planning and control of the movement of an autonomous vehicle must include criteria that are acceptable to humans and, to this end, the needs of the users must be taken into account from the initial stages of its design.

DEVELOPMENT

In order to achieve the integration of the user's emotions and sensations into the behavior of an autonomous vehicle, the SUaaVE project has built a model for evaluating emotions, conceived from a set of rules that determine the emotion felt by a person.

In order to build this model, it has been necessary to collect information about the environment and the passenger of future autonomous vehicles, and to then process and transform this information.

The architecture of the SUaaVE model (Figure 1) is based on a classifier that uses information obtained from an analysis of the environment (weather conditions, traffic conditions...) information about the purposes of the trip (work/leisure trip, arrival time...), the passenger profile and the result of his or her emotional state of mind.

Figure 1. Classification model. From information about the environment to emotion.

More specifically, the classifier works on four different types of information:

• Information about the environment:
Through its sensors, the vehicle registers the information sources provided by the passenger (calendar, personal characteristics and routines) and the external information sources (traffic conditions along the route, weather forecast, etc.).

• The purpose of the trip:
The reasons for the trip (leisure, work, ...) are key factors for the emotion experienced.

• Passenger profile:
The personal characteristics of the passenger are vital. The same stimulus (e.g. a sudden maneuver) can generate different emotions in different individuals, depending on their habits, background, or personal preferences. Therefore, the model includes the individual profiles that influence the emotion for an appropriate classification (age, preferences...).

• Dimensional -emotional driver model:
Based on the driver's physiological information - obtained from sensors inside the vehicle or from cameras - the dimensional model provides information on the intensity of the emotion and whether that intensity is positive or negative.

The classifier is essentially based on an ontology (the structure of knowledge related to emotions) and on an algorithm that extracts from that ontology the emotion candidates that best fit the individual. The output of the reasoner may be one or several emotion candidates.

If there is more than one candidate emotion, the statistical model selects the most likely one. This refinement considers the personal characteristics of the passenger (passenger profile) and the intensity of the emotion (excitation of the dimensional model).

The output of the emotional model is a prediction of the passengers' emotion, classified according to a series of ordered categories (fear, stress, distress, anger, relief, satisfaction). This will allow the modules responsible of decision making as to which maneuvers should be performed in order to effectively manage the emotions of the passengers.

The following tasks have been carried out to develop this model, (Figure 2)

Figure 2. Project phases for the development of the model.

The first step consisted in the generation of a Scenario Universe database through a survey with subjects that was designed to compile potential aspects that could provoke an emotion within the framework of automated vehicles.

 The second step was to analyze and select the most relevant scenarios from the original Scenario Universe database. Between the second and third steps, another survey with subjects was carried out during which we asked participants to rate the selected scenarios.

The third step involved the analysis of the experimental design to extract the main factors influencing emotions.

Finally, the fourth step was the validation of the model. This model will be trained, adjusted and validated with test data collected from the first phase and from the scenario universe database (Figure 3).

Figure 3. Laboratory for the evaluation of the emotional model.

CONCLUSIONS

A first version of the emotional model has been successfully developed to allow us to understand a passenger's emotional state during a journey. This emotional model consists in a categorical model (based on the analysis of contextual factors that influence the on-board travel experience) and a dimensional model (the estimation of the passenger's emotional state as a function of the arousal and valence parameters obtained from physiological signals). Both the categorical and the dimensional models have been designed using a set of algorithms specifically developed by the IBV in the SUaaVE project.

A preliminary laboratory test has allowed us to make an initial evaluation of the dimensional model, obtaining promising results. The next step is to evaluate the model during an initial round of tests with subjects in order to train, validate and make the appropriate adjustments to obtain a reliable emotional state of the passenger on board.

ACKNOWLEDGMENTS

The SUaaVE project is financed by the European Union within the framework of the H2020 research and innovation programme under agreement Nº 814999.