This thesis deals with the animation of virtual characters composed of rigid bodies linked by joints and controlled by physic interactions (forces and torques). The controller is the system that dynamically calculates these interactions.  Our goal is to study and create a controller that is able to control the character interacting with a fluid.

The complexity of the behavior of such environment renders predicting the interactions between the fluid and the character impossible. Therefore, the controller must be able to react to such interactions. We have focused our works on the conception of a SIMBICON typed controller that is able to handle the perturbations caused by the physically simulated fluid. This choice is motivated by our previous controller that was able to handle the interactions with a simplified fluid based on simplified fluid dynamics equations. The use of the complete fluid dynamic equations should allow for a higher realism by taking into account to impact of the character motion on the fluid.

Since our objective is to obtain an interactive controller, we focus our works on two main axes. The first one is the conception of a controller able to handle low simulation frequencies while keeping the high computation speed brought by the use of the SIMBICON framework. We propose to use a feedback system using an online optimization to reduce the instabilities caused by the of the low simulation frequencies. On top of this system, we study the evolution of the value of the PD-controllers depending on the simulation frequency to be able to propose a controller able to handle simulation frequencies as low as 225Hz. The second research axis aims to conceive a fully GPU implementation of a Lagrangian fluid simulation. We study the impact of various optimization proposed by previous work on our GPU implementation. We also present a system allowing us to move the simulation area of the fluid to be able to keep the character centered in the fluid simulation. This system allows us to only use a small area for the fluid simulation allowing us to propose a system showing interactive execution times.