Reference : UMR7253-DOMLOU-007
Nombre de Postes : 1
Workplace : COMPIEGNE
Date of publication : Friday, January 13, 2023
Scientific Responsible name : Domitile Lourdeaux
Type of Contract : PhD Student contract / Thesis offer
Contract Period : 36 months
Start date of the thesis : 1 March 2023
Proportion of work : Full time
Remuneration : 2 135,00 € gross monthly
Description of the thesis topic
Media often highlight the importance of the law enforcement forces in maintaining order (e.g., yellow jackets movement), for health regulations (e.g., COVID19 crisis), or surveillance of the territory in collaboration with the soldiers of the Sentinel Operation (prevention of terrorist attacks). At the same time, they also broadcast violence-related videos and images where some law enforcement officers—despite being trained in avoiding giving way to anger or impulsiveness, keeping safe and adjusted movement coordination, and calm and controlled speech, even in the event of physical or verbal aggression—fail to maintain self-control while facing strong constraints. Therefore, law enforcement executives have found themselves under increasing pressure and scrutiny. Virtual reality technologies could train law enforcement officers, military personnel, health care personnel, and firefighters to cope with stress and regulate their emotions. When facing a violent / aggressive / menacing individual or a panicked victim, the learner will be trained to maintain a high level of self-control and attention, to slow down the sequence of events, enhance situational awareness, conduct proper threat assessments, and select and produce appropriate verbal and nonverbal behaviors. The objective of this PhD thesis is to generate automatically virtual personalized scenarios that consider the trainees' cognitive profiles.
We aim for the reification of cognitive dissonance . The theory of cognitive dissonance is an internal tension inherent in a person's system of thoughts, beliefs, emotions, and attitudes (cognitions) when several of them contradict each other. We want to select situations that oppose several value systems (springing from professional deontology, state morality, cultural beliefs, personal ethics, etc.). The goal is to push the learner into cognitive dissonance bordering on losing control. Among situations of cognitive dissonance, dilemmas represent an interesting challenge. Indeed, dilemma situations are difficult to represent in classical logic and have been little studied. We can rely on the work of  on the generation of dilemmas as well as on the theory of universal value of  to establish a moral profile. We will design a computational narrative model based on the reification of cognitive dissonance theory to generate dissonance situations and dilemmas. To offer customizable scenario, writing scenarios requires significant work. This effort necessary for the scaling up coherent and precisely controlled scenarios is called the authoring bottleneck. Then, we aim for an orchestration system capable of generating adaptive environments, without having to explicitly define all the possible scenarios. To preserve the freedom of action and to ensure the adaptivity of the behaviors, we assume that it is possible to dynamically and automatically generate training situations from knowledge models that underlie the simulation (adapting the type of crowd, interactions with protagonists, aggressivity, hostility, situated elements, etc.). It is impossible to script everything. Thus, we want to design an orchestration system that is sufficiently generative to offer a wide range of situations and to be able to re-generate others if the learner goes against the initial scenario. We aim to produce as many combinations of narrative objectives and situations as the existing profiles, considering that the profiles are dynamically generated and that there is a wide variety of profiles. We hypothesize that the generative power of the system will allow us to achieve the objectives of variability and resilience. We need to generate observables and multicriteria measurements to dynamically update the user profile. This is a constraint satisfaction problem (CSP). To tackle this issue, we will propose a hybrid approach combining 1) a numerical approach with probabilistic graphical models to respect the authorial intention and with belief functions to control the situations, 2) a semantic approach with a knowledge model to improve expressiveness and to respect the coherence, 3) a logical model and planning techniques to generate robust and resilient scenarios.
 Festinger L. A theory of cognitive dissonance. Vol. 2. Stanford university press, 1957
 Benabbou A, Lourdeaux D, and Lenne D. Automated dilemmas generation in simulations. In: Cognition, Technology & Work 23.1. pp. 161–175
 Schwartz SH, Cieciuch J, Vecchione M, Davidov E, Fischer R, Beierlein C, Ramos A. Refining the theory of basic individual values. Journal of personality and social psychology 103, no. 4: 663. 2012
The Heudiasyc Laboratory was created in 1981. From the start Heudiasyc has been closely allied to CNRS and is attached to CNRS's INS2I (Information Sciences) section. Heudiasyc's research is in the field of information and digital technology (computer science, automatic control, robotics, and artificial intelligence).
The aim is to develop ways of representing, analysing and controlling systems that are subject to criteria and constraints, whether these be expressed in scientific, technological, economic, or social terms. Research is organized around three teams:
- CID: Knowledge, Uncertainty, Data
- SCOP: Dependability, Communication, Optimization
- SyRI: Robotic systems in interaction
The thesis will take place in the CID team.
We talk about it on Twitter!