A multi-modal and multi-criterion serious game to support the rail eco-driving based on human behavior learning
Résumé
Eco-driving is an important issue for transportation system and more precisely for rail domain. The main difficulty of such development concerns the real-time constraints based on the optimum research problem. Different criteria such as the safety, the time schedule, or the human vigilance can be combined in order solve it and define a setpoint to optimize the consumption criterion. This paper proposes a serious game for defining an optimal eco-driving profile that respects the time schedule and the safety. Different kinds of rail system are proposed with one circuit: a high speed train, an urban train and a tram. The serious game integrates three models to define an optimal eco-driving instruction: the train model, the consumption model and the human behavior model. The train model consists in determining the acceleration and deceleration constraints regarding the position of the manipulator that aims at stopping, braking or moving the train. The consumption model integrates the electrical power and the train speed. The human behavior model consists in copying previous behaviors at a given time for defining the optimal eco-driving instructions at the next time. It is based on the concept of the mirror effects implemented into a new original Learning system. This system aims at learning from the human behavior and its impact on the consumption, the time schedule and the safety. Regarding previous data on these criteria, it researches similar data vector on a knowledge base in order to propose the corresponding optimal eco-driving instructions for the manipulator position at the current time. The learning process consists in reinforcing the content of the knowledge base by applying mirror effects. Two Mirror effects are implemented. The former effect limits the number of cases into the knowledge base and a new case is compared to the existing one in order to find the winner case that is closer to the new one and to determine the new corresponding eco-driving instructions. The reinforcement process consists in mirroring the differences between new input data and the winner case into the entire knowledge base. The reinforcement process of the later effect mirrors all new data into the knowledge base by considering them as new cases and by assessing the optimal eco-driving profile. The serious game proposes an interface to select the simulated vehicle and the mode of interaction, i.e., a keyboard, a joystick, pedals or the gamer’s voice. The gamers must drive train into a predefined circuit by respecting an eco-driving instruction that informs them about the position of the manipulator to be achieved in order to optimize the electrical consumption and to respect the time schedule and the safety constraints (acceleration, speed limitation). The final paper will present some results obtained by students who have been awareness about energy consumption problems by plying with the proposed multi-modal and multi-criterion serious game.