Source-linked AI summary
Projective Affine Body Dynamics for Multibody Systems
Zimeng Ye, Xiaowei He, Yuzhong Guo, Yin Yang, Chenfanfu Jiang, Hongan Wang
TL;DR
Multibody dynamics must handle nonlinear constraints, non-smooth contact, and flexible-body behavior while remaining efficient to parallelize. The paper reformulates the problem using affine bodies and peridynamic bonds, then introduces a Hessian-free GPU solver that supports nonlinear joints, contact, and friction. It reports slightly better hard-constraint enforcement than AVBD in one comparison, while retaining limitations under extreme rotations and for higher-order barriers or fluids and deformable objects.
Problem
Non-smooth contact, nonlinear flexible-body constraints, and changing contact or constraint sparsity make constrained multibody dynamics difficult to model and parallelize efficiently.
Method
The method reformulates constrained multibody dynamics variationally as affine bodies connected by peridynamic bonds and solves nonlinear terms with semi-implicit successive substitution in a Hessian-free projective framework.
Results
The method supports real-time multibody dynamics with nonlinear joint constraints, contact, and friction, and is slightly better at enforcing hard constraints in one comparison.
Takeaways & Limitations
Modeling constraints and contact within the affine-body framework provides a unified basis for highly parallel GPU simulation of rigid and flexible interactions.
Takeaways & Limitations
Simulations can fail when successive-iteration rotations are large, higher-order polynomial barriers become increasingly complex, and extension to fluids and deformable objects remains unclear.
Abstract
from arXiv · showhide
Multibody systems have widespread applications in diverse fields such as robotics, entertainment, and animation. Their capability to model and simulate intricate interactions among interconnected bodies makes them invaluable in various domains. However, complexity arises with factors like non-smooth contact handling, nonlinearity in flexible joints, and parallelization challenges. We introduce a stable and highly parallel GPU algorithm within affine body dynamics for solving constrained multibody dynamics with nonlinear constraints.Our innovation involves reformulating constrained multibody dynamics into a variational form, treating the system as a set of affine bodies connected with peridynamic bonds. This formulation provides a unified model for affine body dynamics, constraints, and contact within the peridynamics framework.It also facilitates the integration of the semi-implicit successive substitution method to solve nonlinear optimization in a global-local iterative manner. The proposed method obviates the necessity of assembling a global Hessian, rendering it highly suitable for efficient implementation on GPUs.This allows real-time simulation of intricate interactions involving both rigid and flexible bodies, encompassing contact, joint constraints, and friction.
1. Introduction
Constrained multibody dynamics becomes difficult with non-smooth contact, nonlinear flexible-body behavior, and GPU parallelization. The paper proposes a GPU-oriented affine-body formulation with a Hessian-free projective solver.
- Non-smooth contact, nonlinear flexible bodies, and parallelization create significant challenges for multibody dynamics.
- Conventional formulations typically derive constraint forces and reduce the problem to a Linear Complementarity Problem, but this does not adequately model non-smooth and nonlinear contact or joint constraints.
- The proposed method models multibody systems as affine bodies with rigidity constraints and uses a projective local/global alternating solver.
- Constraints are represented as peridynamic bonds, enabling the semi-implicit successive substitution method to address nonlinear terms in the global step.
- The contributions include a Hessian-free projective optimizer, an inverse for a 12 × 12 block diagonal matrix, and angular-velocity recovery for rigid bodies.
2. Related Work
Related work develops computational approaches for interconnected rigid and flexible bodies, including contact, friction, projective dynamics, and affine-body formulations. These methods improve robustness and efficiency but retain limitations in material generality and parameter sensitivity.
- Multibody dynamics analyzes interconnected rigid or flexible bodies, with challenges including representation, computational cost, stability at large time steps, joints, contacts, and friction.
- Rigid-body contact methods introduced impulse handling, time integration, shock propagation, and contact models using mass, location, and velocity information.
- Incremental potential contact guarantees intersection-free configurations, while affine body dynamics unifies simulations of stiff and soft materials.
- Position-based and projective dynamics extend to elasticity, contact, friction, and fluid incompressibility, but can be sensitive to time-step and iteration-count parameters.
- Projective-dynamics research recasts implicit integration as energy minimization, enables prefactored global solves, accelerates convergence, and studies quasi-Newton formulations.
- General hyperelastic-material modeling remains elusive within projective dynamics, motivating semi-implicit successive substitution approaches.
3. Foundations
Affine-body foundations describe each body through a time-varying linear transform and translation, with an orthogonality potential restoring rigidity. Independent bodies yield GPU-parallel block-diagonal Hessians, whereas contacts and constraints create difficult off-diagonal sparsity.
- An affine body is described by a time-varying linear transform A(t) ∈ R3×3 and translation c(t) ∈ R3.
- Material points are mapped using body-frame material coordinates, and the body Jacobian uses I3 and I3 ⊗ xiξ.
- Figure 2 demonstrates mechanisms combining multiple joints, contact, and friction.
- A stiff orthogonality potential restores rigid-body rotational degrees of freedom through energy minimization, with κ controlling rigidity.
- Without contacts or constraints, the global Hessian is a 12 × 12 block diagonal matrix that can be fully parallelized on GPUs.
- Contacts and constraints introduce unpredictable off-diagonal Hessian terms, motivating a matrix-free solver with memory cost linear in the number of degrees of freedom.
4. Projective Affine Body Dynamics
The section introduces the optimization problem for multibody dynamics and provides notation resources for the subsequent projective affine body dynamics formulation.
- The paper first revisits the optimization problem for multibody dynamics before presenting its solution.
- Table 1 summarizes the notation used in the following discussion.
4.1. Motivation
The method reformulates constrained multibody dynamics using affine bodies connected by peridynamic bonds, linking rigid-body and deformable-body descriptions through variational energy minimization.
- The system models rigid bodies connected by joints that restrict relative motion through zero to six constraints.
- Affine bodies connected by peridynamic bonds impose additional forces and convert the equations of motion into an optimization problem.
- Bond energies and concentrated forces provide the variational and force-based relationships used to express dynamics in affine coordinates.
- Polar decomposition relates affine transformations to rotations, while the affine moment of inertia updates from its material-space value.
4.2. Projective Semi-implicit Solver
The solver alternates a global semi-implicit substitution step with a local rigidity projection, producing independent body updates while addressing nonlinear bond energies.
- Global Solve: Bond energies depend on inter-body distance, and their derivatives may contain nonlinear terms that make Newton-type methods expensive.
- Global Solve: The global step temporarily ignores rigidity constraints and applies semi-implicit successive substitution to the nonlinear optimization problem.
- Local Solve: The local step rigidifies each affine body by projecting the global iterate onto a feasible solution before returning it to the next global computation.
- The global and local steps alternate until a user-defined threshold or the maximum iteration count is reached.
- Global Solve: The global formulation becomes a fixed-point iteration with one separable 12 × 12 matrix per affine body, enabling independent GPU updates.
- Local Solve: Increasing κ makes affine bodies progressively stiffer, but extremely large κ creates ill-conditioning and can prevent a unique minimizer.
4.3. A sparse representation for Hi
The sparse representation exploits the structure of each 12 × 12 matrix H_i to reduce GPU storage and simplify inversion.
- Dense storage of H_i can exhaust GPU registers, while direct LU inversion of a 12 × 12 matrix may require at least 144 registers.
- The method exploits the sparse structure of J to simplify H_i into scalar, vector, and matrix blocks.
- Storing a_i, b_i, and D_i instead of the full matrix reduces the required register count from 144 to 13.
- The structured inverse retains the same form as H_i, simplifying inverse computation and reducing register requirements for GPU execution.
4.4. Update translational and angular velocities of rigid bodies
After optimization, translational velocities are updated directly, while angular velocities are recovered from the affine transformation through a skew-symmetric matrix relation.
- Translational velocities are updated from the optimization solution using the standard position-based dynamics procedure.
- Angular velocity components are derived from the off-diagonal entries of [ω]× = ȦA^-1 and averaged to reduce floating-point errors.
- The complete procedure is summarized in Algorithm 1, with rotational components obtained through standard quaternion-to-matrix conversion.
5. Results and Discussion
The method supports real-time multibody simulations involving joints, contact, friction, and complex mechanisms. It remains stable for large rigid-body systems and achieves more than an order-of-magnitude speedup over the Newton barrier method.
- The implementation uses C++ and CUDA on an NVIDIA RTX 3080 GPU, with linear BVH collision detection and Coulomb friction.
- Joints: Soft and hard constraints are combined to model spherical, hinge, and slider joints.
- Joints: The method simulates mechanisms with multiple rigid bodies, hinge joints, slider joints, driven gears, contact, and friction.
- Contact and Friction: The method stably solves contact and friction for a windmill colliding with two piles of boxes.
- Contact and Friction: The ball-and-boxes example produces consistent results compared with the sequential impulse method.
- More than one order of magnitude speedup is achieved compared with the Newton barrier method.
5.2. Comparison to other methods
The comparisons examine convergence, hard-constraint enforcement, and behavior under large mass ratios. The method converges similarly or slightly faster in tested spring systems, slightly improves hard-constraint enforcement, and requires augmented Lagrangian treatment for extreme mass-ratio chains.
- Comparison to other methods: Against the sequential impulse method, the proposed method runs at approximately 40 fps versus approximately 70 fps.
- Comparison to the AVBD method: For a single stretched spring, AVBD reaches the exact solution in one iteration, whereas the proposed method has linear convergence and needs multiple iterations.
- Comparison to the AVBD method: In an 11 × 11 box grid with 220 springs, the proposed method converges slightly faster initially, while both methods eventually reach similar results.
- Comparison to the AVBD method: The proposed method is slightly better than AVBD at enforcing hard constraints.
- Comparison to the AVBD method: Without augmented Lagrangian formulation, large mass ratios impair distance preservation; with it, results become comparable to AVBD.
5.3. More examples
The examples demonstrate the method across large-scale rigid-body scenes, deformable coupling, and constrained systems, while comparisons examine convergence and settling behavior against AVBD.
- Bridge: Our method converges slightly faster during early iterations than AVBD on the spring-grid scenario, but their eventual convergence behavior is similar.For a single rigid box, AVBD converges faster; the relative error follows the definition in [BML∗14].
- Bridge: An 11×11 grid of boxes connected by 220 springs forms the nonlinear spring-constraint scenario used for convergence comparison.Each box connects to four neighbors, with springs initially pre-stretched toward the scene’s middle-bottom.
- More examples: The method supports efficient real-time simulation of rigid-deformable coupling, including a sailboat interacting with Gerstner waves and shallow-water wake dynamics.The boat–ocean interaction uses weak coupling, with buoyancy computed directly.
- Large-scale stacking and collision: Large-scale collision examples include balls impacting 50 × 50 × 50 block piles, simulated at approximately 5 fps without complete performance optimization.
- More examples: Augmented-PABD performs slightly better than AVBD at settling a chain of rigid boxes with hard constraints.
- More examples: With a large mass ratio, our method produces results comparable to AVBD, whereas omitting the augmented Lagrangian fails to enforce distance constraints well.
6. Conclusion
The paper develops projective affine body dynamics as a stable, highly parallel GPU algorithm for constrained multibody dynamics with nonlinear constraints. Its formulation supports a Hessian-free real-time solver, while the authors identify failure under extreme rotations and unresolved extensions to more complex fluid and deformable scenarios.
- Projective affine body dynamics reformulates constrained multibody dynamics variationally, representing the system as affine bodies connected by peridynamic bonds.
- The formulation integrates semi-implicit successive substitution for nonlinear global-step terms and avoids assembling a global Hessian.
- The method targets real-time multibody simulation with nonlinear joint constraints, contact, and friction on highly parallel GPU architectures.
- Simulations can fail under extreme conditions, including large rotations between successive iterations.
- Higher-order polynomial barrier functions make positive-and-negative energy decomposition increasingly complex, and extensions to fluids and deformable objects remain unclear.