Project : alcove
Section: Scientific Foundations
Interaction in complex virtual environments
The interaction in complex virtual worlds covers two main problems: collision detection and response between virtual objets on the first hand, and real time intervention of the user into the simulation eventually via haptic devices on the second hand 
Interaction between virtual objects in complex environments
For many years now, we have been investigating the use of spheres approximating the volume of virtual objets: Spheres are very attractive, since it is straightforward to check their intersection and compute a penalty-based collision response. However, they have revealed some major drawbacks : they fail to represent surface body properly, they sometimes poorly approximate the volume of certain virtual bodies and in case of rapid motions, they do not allow an efficient and physically correct computation of the collision response. During his Master Thesis  , D. Marchal has investigated other real-time approaches for collision detection and penalty computation. They are mainly based on depth map computed for each objects, and updated after any deformation.
In the continuation of his Master Thesis, Damien Marchal started a PhD about collision problems (detection, response, self-collisions, generic collision models) in complex virtual environment.
The interactions between virtual objects also include the links between objects, and more generally any bi-lateral constraints. To handle these constraints in real-time, we have investigated different existing approaches . The projection method is more straightforward and fast, but it hardly deals with multiple constraints on the same object and tends to perturbate the resolution of the physical equations too much. Some displacement constraints have also been investigated but the more interesting results have been obtained with lagrangian multipliers. However, in our simulator, this implies that these constrained bodies should be encapsultated in a unique "articulated" object. Models must therefore be designed in a compatible fashion and that is why only few objects of our simulator can be included in articulated structure at the present time.
Interaction via haptic devices
We have studied how to use the proxy/god-object method for the control of certain virtual bodies of the simulation. Our approach consists in considering the proxy as a simulated body (as any other simulated object of the environment) which must reach the actual position of the device. Several problems do appear however. Due to their "dynamic" nature, proxies exhibit a high inertia and the manipulated objects react very slowly and can oscillate around a rest position before beeing motionless.
We have tried to use only a first order equation to determine the position of the proxy (and get what we call "a cinematic body"), but this approach is less stable. This inertia effect must however be adressed because it induces high feedback forces which make the user think the manipulated object is "heavy" (it is not : it is only "slow"). This problem can be solved by computing a critical damping, but this implies that the simulation is computed at a high frequency (1Khz) . Further investigation is needed, but this approach is more appropriate to avoid large interpenetration of manipulated objet in the environnement. >From a point of view, the manipulated objets are controled in an admittance fashion instead of a classical impedance fashion: that is, we send force to the proxy and get a position as a result. This makes the use of admittance devices such as the Virtuose easily plugable in our simulator.
S. Fonteneau is currently investigating the use of virtual proxies to manipulate more complex tools than a simple 6 Dof rigid body : the use of articulated-rigid bodies is necessary to control surgical forceps for instance.