Adaptive Autonomy for a Human-Robot Architecture

In the context of supervisory control of one or several artificial agents by a human operator, the definition of the autonomy of an agent remains a major challenge. When the mission is critical within a dynamic environment, e.g. in the case of uninhabited vehicles, errors are not permitted while performance must be as high as possible. Therefore a trade-off must be found between manual control, usually ensuring good confidence in the system but putting a high workload on the operator, and full autonomy of the agents, often leading to less reliability in uncertain environments and lower performance. Having an operator in the decision loop does not always grant maximal performance and safety anyway, as human beings are fallible. Additionally when an agent and a human decide and act simultaneously using the same resources, conflicts are likely to occur and coordination between these heterogenous entities is mandatory. We present the basic concepts of an approach that will aim at dynamically adjusting the autonomy of an agent in a mission relatively to its operator, based on a formal modelling of mission ingredients