A great challenge for interactive virtual characters is to be able to interact with human operators by performing tasks in physics-based virtual environments. This dissertation is particularly interested in the interaction with operators who are slightly immersed, that is to say, with operators having the minimum equipment necessary for the interaction, for example, a simple motion capture of the hands. This involves endowing the virtual character with the ability to adjust its postures autonomously, and to accomplish tasks required by operators in real time by trying to follow their motions as best as it can, while autonomously handling multiple constraints due to interactions with virtual environments. In this dissertation, we present an original hybrid control system that allows us to realize an interactive virtual character with certain levels of autonomy. We propose a posture optimization approach, which allows the virtual character to search for optimal and robust postures, including contact positions, before performing a given manipulation task. We develop a multi-objective control framework, which can handle multiple task objectives and multiple unilateral and bilateral contacts with or without friction. It allows the character to perform motion tracking and object manipulation tasks in a physics-based virtual environment, while interacting with an operator in real-time. We develop a wrench-bound method. It is a novel prioritized control, which allows inequality constraints on the higher-priority task, and can ensure the passivity of the system to guarantee stable operations.
