Frictional Contact in Interactive Deformable Environments

The use of simulations provides great advantages in term of economy, realism, and adaptability to user requirements in many research and technological fields. For this reason simulations are currently exploited, for example, in prototyping of machinery parts, in assembly-disassembly test or training and, recently, simulations have also allowed the development of many useful and promising tools for the assistance and learning of surgical procedures. This is particularly true for laparoscopic intervention. Laparoscopy, in fact, represents the gold standard for many surgical procedures. The principal difference from standard surgery is the reduction of the surgeon ability to perceive the surgical scenario, both from visual and tactile point of view. This represents a great limitation for surgeons who undergo long training before being able to perform laparoscopic intervention with proficiency. This, on the other hand, makes laparoscopy an excellent candidate for the use of simulations for training. Some commercial training softwares are already available on the market, but they are usually based on rigid body models that completely lack the physical realism. The introduction of deformable models may leads to a great increment in terms of realism and accuracy. And, in the case of laparoscopy trainer it may allow the user to learn not only basic motor skills, but also higher level capabilities and knowledge. Rigid bodies, in fact, represents a good approximation of reality only in some situations and in very restricted ranges of solicitations. In particular, when non engineering materials are involved, as happens in surgical simulations, deformations cannot be neglected without completely loosing the realism of the environment. The use of deformable models, however, is limited for the high computational costs involved in the computation of the physics undergoing the deformations and because of the reduction in pre computable data in particular for collision detection between bodies. This represents a very limiting factor in interactive environments where, to allow the user to interactively control the virtual bodies, the simulation should be performed in real time. In this thesis we address the simulation of interactive environment populated with deformable models that interact with frictional contacts. This includes the analysis and the development of different techniques which implement the various parts of the simulation: mainly the methods for the simulation of deformable models, the collision detection and collision solution techniques but also the modelling and the integration of suitable friction models in the simulation. In particular we evaluated the principal methods that represent the state of the art in soft tissue modeling. Our analysis is based on the physical background of each method and thus on its realism in terms of deformations that the method can mimic and on the ease of use (i.e. method understanding, calibration and ability to adapt to different scenarios) but we also compared the computational complexity of different models, as it represents an extremely important factor in the choice and in the use of models in simulations. The comparison of different features in analyzed methods motivated us to the development of an innovative method to wrap in a common representation framework different methodologies of soft tissue simulation. This framework has the advantage of providing a unified interface for all the deformable models and thus it provides the ability to switch between deformable model keeping unchanged all other data structures and methods of the simulation. The use of this unique interface allows us to use one single method to perform the collision detection phase for all the analyzed deformable models, this greatly helped during the identification of requirements and features of such software module. Collision detection phase, when applied to rigid bodies, usually takes advantage of pre computation to subdivide body shapes in convex elements or to construct partitions of the space in which the body is defined to speed up the computation. When handling deformable models this is not possible because of the continuous changes in bodies shape. The collision detection method used in this work takes into account this problem and regularly adapt the data structures to the body configuration. After collisions have been detected and contact points have been identified on colliding bodies, it is necessary to solve the collision in a physics based way. To this extent we have to ensure that objects never compenetrate during the simulation and that, when solving collisions, all the physical phenomena involved in the contact of real bodies are taken into account: this include the elastic response of bodies during the contact and the frictional force exerted between each pair of colliding bodies. The innovative method for solving collision that we describe in this thesis ensures the realism of the simulation and the seamless interaction with the common framework used to integrate deformable models. One important feature of biologic tissues is their anisotropic behavior that usually comes from the fibrous structure of these tissues. In this thesis we propose a new method to introduce anisotropy in mass spring model. The method has the advantages of preserving the speed and ease of implementation of the model and it effectively introduces differentiation of the model behavior along the chosen directions. The described techniques have been integrated in two applications that allows the physical simulation of environments populated with deformable models. The first application implements all the described methods to simulate deformable models, it performs precise collision detection and solution with the possibility to chose the most suitable friction model for the simulation. It demonstrates the effectiveness of the proposed framework. The main limitation of this simulator, i.e. its high computation time, is tackled and solved in a second application that exploits the intrinsic parallelism of physical simulations to optimize the implementation and to exploit parallel architecture computational power. To obtain the performances required for an interactive environment the simulation is based on a simplified collision detection algorithm, but it features all the other techniques described in this thesis. The parallel implementation exploits graphic cards processor, a highly parallel architecture that update the scene every milliseconds. This allows the rendering of smooth haptic feedback to the user and ensures the realism of the physics simulation. The implemented applications prove the feasibility of the simulation of complex interactions between deformable models with physics realism. In addition, the parallel implementation of the simulator represents a promising starting point for the development of interactive simulations that can be used in different fields of research, such as surgeon training or fast prototyping.