Subscribe to RSS - multi agent simulation

multi agent simulation

Funded by the French Ministry of Education and Research and the French National Research Agency (ANR-17-CE22-0010)

Autonomous vehicles typically function well on large roads i.e. in structured and predictable environments. However and increasing trend in urban design is to promote shared spaces where segregation between pedestrains and vehicles is minimised by removing features such as kerbs, road markings, traffic signs and traffic lights. The context of the work is to focus on safe and predictable interaction between cyber-cars and other road users in complex and human populated urban environments (i.e. shared spaces). A social and cognitive dimension is required in the intelligent vehicles so that they will follow socially adapted behaviors that will be understood and predicted by passengers and pedestrians; will communicate their intentions to move or stop; and will increase safety. The vehicles will not simply stop during uncertain situations but will be proactive and negotiate their paths and velocities, by conveying their course and direction of travel to surrounding road users.

Models of human and crowd behavior will be built in order to give the autonomous car knowledge of its social environment in the present and the near future. These models will integrate knowledge from social sciences (proxemics, behaviorism) and will use agent based simulation. The trajectories of the autonomous vehicle are expected to be safe and socially compliant and the decisions of the vehicle will be based on the estimated intentions of all agents of the scene.