Objective: One aspect to be taken into account is the dynamic adaptation of content to the learner's behaviour. In work situations involving risks, and in highly degraded situations, there is clear value in allowing users or learners, depending on their level and on the targeted training objectives, to interact freely with the system and to make mistakes so that they can see their impact. To allow such interactions and the emergence of knowledge, and hence of learning, the complex functioning of these human and technical systems must be modelled and scenarised in relation to the targeted objectives. The aim is therefore to steer the scenario of a virtual training environment towards relevant situations. To generate such situations and provide material for discussion, it is useful to be able to explain why these situations occur, and to provide users with elements for assessment, or even self-assessment, so that they can understand the behaviours or events that led to them. In particular, in highly complex situations involving human, technical and organisational factors, with diverse and sometimes delayed causes, a tracing system helps to analyse situations more effectively. Such systems enable trainers to monitor learners' activity, and support reflective activities for users or learners.

Motivations:

• Increase the efficiency and quality of simulations, in order to offer complex simulations that are relevant from a learning perspective
• Allow learners to experience highly interactive situations (situated learning through error) that are relevant for learning and consistent in content
• Learn to react quickly and appropriately in degraded, stressful or rare situations
• Offer situations with no ideal solution, and discuss the benefits and risks of possible solutions (trade-off situations)
• Help learners understand the cause-and-effect links between actions and the occurrence of critical events in highly complex systems, where these links are not always obvious
• Confront learners with situations in which their professional obligations conflict with their own moral values, in order to train decision making under emotional tension

Proposed approach: Our work focuses on the dynamic adaptation of the scenario, intelligently orchestrated according to the assessment of the situation. We propose complex scenarios and simulations that are open but directed by:

• mechanisms controlling interactions and complexity based on the diagnosis of the learner's behaviour (errors, successes, history, etc.), their profile (level, preferences, etc.) and the learning objectives;
• scenario and dramaturgical rules that manage dramatic intensity and increase the learner's emotional involvement in learning (dramatic tension, pacing, complexity of the situation, etc.).

In addition, we propose learner and trace monitoring to support learners' reflective activities and trainers' analysis;

• the generation of moral conflict situations, in which learners must choose between their professional moral obligations and moral values selected according to their profile and vulnerabilities (Jeanne Parisse, MASTERS project).

Novelty:

• Dynamic adaptation of the scenario and of the behaviour of autonomous virtual characters
• Controlled freedom of interaction
• Regulation of the complexity and severity of learning situations (profile, history, learning situations, activity)
• Sense of achievement while raising awareness of consequences
• Tensions and emergency situations (identified through professional didactics)
• Generation of moral conflict situations adapted to the learner's vulnerabilities
• Hybrid generation combining ontologies, fuzzy cognitive maps and large language models, ensuring both variability and verifiability
• Activity description using the ACTIVITY-DL formalism (formerly HAWAI-DL)
• Traces of highly complex activities