A position-based dynamics system for animated character effects

Abstract

Figure 1: (left) Typical animated scene containing multiple simulated elements including hair, cloth, cargo nets and ropes; (right) complex garment model for the character Stoick that mixes cloth, armor, leather, and even his beard in a single interacting cloth model that executes in 0.3 secs per frame on an artist desktop.During production of How To Train Your Dragon 2 we were motivated to improve our technology for character effects such as ropes, chains and cloth. Our goal was to reduce artist iteration time for simulation work while retaining the flexibility required for art direction. Inspired by the success of position-based dynamics in games, we implemented a new simulation pipeline for deformable solids based on a fast, stable simulation engine that has delivered significant improvements. SimulationPosition-based dynamics is attractive because it produces algorithms that are fast, unconditionally stable, and controllable. The technology is well established in games, virtual reality and special effects where it is used extensively for rigid body dynamics [Bender et al. 2013]. However, position-based solvers are not widely used in film production for deformable solids, and there is little in the literature to indicate the challenges in adopting these methods for feature animation. In this talk we introduce a system based on the Carbon dynamics library with novel features for performance, setup flexibility, art directability and control.Performance is addressed by using parallel nested collision detection that enables aggressive multicore scaling. Parallelism is implemented using Intel Threading Building Blocks. The algorithm is based on order-deterministic dynamic task splitting and recycling. We process collision detection for the entire scene by starting from the broad phase and nesting the narrow phase and contact generation. We show that this approach generates excellent scaling for typical workloads on up to 16 cores. Setup flexibility is addressed by solving a single heterogeneous system for the interaction of all soft and rigid bodies. Collision constraints and joint constraints between dynamic or kinematic, rigid or deformable objects are handled within a single solver. All object collisions can be modeled using a common set of geometric primitives: points, edges, and single or double sided faces. All primitives may have a user-specified thickness. Faces can be arranged with no requirements for convex decomposition. This heterogeneous systems supports complex cloth models, e.g. mixed soft and rigid bodies for garments that combine elements like cotton and leather that interact within a single simulation.Art direction for motion is supported via animated position constraints that drive a simulation towards an animated goal shape, while respecting other constraints such as collision surfaces. Artists may also animate the rest shape of a model in order to control effects such as squash and stretch, which require the simulation to adapt to non-physical changes. We deal with pinching, another form of no...

Data‐Driven Facial Simulation

“…XPBD relies on an iterative solver based on the Gauss–Siedel method and uses compliant constraints which achieve excellent results in few iterations. XPBD approaches are rapidly becoming the method of choice for simulation in the creative industry [vel, SLG14].…”

Muscle and Fascia Simulation with Extended Position Based Dynamics

“…PBD is making its entry into the animation industry [SLG14], but prior to the work from Romeo et al . [RMSQ18] there was little literature about muscle or fascia simulation with PBD [Fra12] and no reference to any solution that integrates fibres' activation into the simulation loop.…”