Physically Accurate Rigid-Body Dynamics in Particle-Based Simulation
Ava Abderezaei*, Nataliya Nechyporenko, Joseph Miceli, Gilberto Briscoe-Martinez, Alessandro Roncone · 2026
Read Physically Accurate Rigid-Body Dynamics in Particle-Based Simulation by Ava Abderezaei*, Nataliya Nechyporenko, Joseph Miceli, Gilberto Briscoe-Martinez, Alessandro Roncone (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14634.txt.
Introduction
Abderezaei and co-authors make particle-based simulation respect rigid-body physics with PBD-R. Position-based dynamics is fast and looks right, but its physics can drift. Robotics needs the accuracy that stitched multi-solver tools strain to give. PBD-R adds a momentum-conservation constraint to the classic solver. A revised velocity update completes the fix. A solver-free benchmark with known solutions grades physical accuracy. PBD-R beats plain PBD by a wide margin there. It rivals MuJoCo accuracy at lower compute cost, they report.
Catalog record
| CREATOR | Ava Abderezaei*, Nataliya Nechyporenko, Joseph Miceli, Gilberto Briscoe-Martinez, Alessandro Roncone |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42100 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.14634
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Physically Accurate Rigid-Body Dynamics in Particle-Based Simulation Ava Abderezaei*, Nataliya Nechyporenko, Joseph Miceli, Gilberto Briscoe-Martinez, Alessandro Roncone Abstract Robotics demands simulation that can reason about the diversity of real-world physical interactions, from rigid to deformable objects and fluids. Current simulators address this by stitching together multiple subsolvers for different material types, resulting in a compositional architecture that complicates physical reasoning. Particle-based simulators offer a compelling alternative, representing all materials through a single unified formulation that enables seamless cross-material interactions. Among particle-based simulators, position-based dynamics (PBD) is a popular solver known for its computational efficiency and visual plausibility. However, its lack of physical accuracy has limited its adoption in robotics. To leverage the benefits of particle-based solvers while meeting the physical fidelity demands of robotics, we introduce PBD-R, a revised PBD formulation that enforces physically accurate rigid-body dynamics through a novel momentum-conservation constraint and a modified velocity update. Additionally, we introduce a solver-agnostic benchmark with analytical solutions to evaluate physical accuracy. Using this benchmark, we show that PBD-R significantly outperforms PBD and achieves competitive accuracy with MuJoCo while requiring less computation. I Introduction Simulation is central to modern robotics, underpinning policy learning, model-based control, planning, and sim-to-real transfer. Robots rely on simulation to reason about how their actions impact the world: for example, pushing a container of liquid requires anticipating both the container’s motion and the fluid’s dynamics. However, the typical simulation stack is fragmented in how it models diverse materials. It uses separate engines that are “best” for specific physical properties: rigid-body engines for rigid objects, finite element methods for deformables, and smoothed particle hydrodynamics for fluids [ 21 ] . As a result, simulators have to rely on heterogeneous subsolvers to model complex environments. This fragmentation has significant downstream consequences. It forces each material to rely on specialized contact models, complicating cross-material interaction and breaking the end-to-end differentiability required for gradient-based optimization and learning. Errors compound across incompatible subsolvers, making simulation pipelines brittle and difficult to scale to the contact-rich manipulation tasks robots must perform in the real world. Fig. 1 : To leverage the benefits of particle-based simulation for robotics, we introduce PBD-R, a revised Position-Based Dynamics formulation that addresses the rigid-body physical accuracy gap in standard PBD. Contribution 1 (Top): PBD-R introduces modifications to PBD to achieve physical accuracy. Contribution 2 (Right): We introduce a solver-agnostic benchmark of seven physics tests with analytical solutions for evaluating simulator accuracy. Contribution 3 (Bottom): PBD-R outperforms PBD and achieves competitive accuracy to MuJoCo while requiring less computation. Within this fragmented landscape, particle-based representations have emerged as a compelling alternative. Unlike the mesh-based pipelines that underpin the majority of existing simulators, particle-based methods offer a fundamentally more expressive and unified representation for physical simulation. By treating rigid, deformable, cloth, and fluid entities as a set of particles, a single contact model and solver can handle all interactions uniformly [ 14 ] . Beyond unified simulation, particle-based representations provide additional advantages due to their meshless formulation: They support direct collision checking without requiring convex decomposition of complex geometry [ 25 ] , enable faster collision computation through simplified proximity queries [ 20 ] , and permit straightforward digital twin construction by mapping from point clouds to simulation-ready particle sets [ 2 ] . Position-Based Dynamics (PBD) is a particularly attractive particle-based framework due to its computational efficiency, numerical stability, and support for a wide range of constraint types [ 14 , 15 , 5 ] . A recent variant of PBD [ 22 ] , further enables differentiable simulation. It has also seen increasing adoption in robotics for tasks such as cloth manipulation [ 26 ] and digital twin construction [ 2 ] . However, PBD was originally developed for computer graphics, where computational efficiency and visual plausibility are prioritized over physical accuracy [ 5 ] . This is a critical limitation for robotics, where physical fidelity directly determines the sim-to-real gap [ 4 ] and the downstream viability of policy transfer to the real world. Reducing this gap requires simulators that model physical dynamics with high accuracy. To this end, we introduce PBD-R, which improves the physical accuracy within the unified particle representation paradigm of PBD. PBD-R improves the physical accuracy of rigid-body dynamics without sacrificing the benefits of the particle-based framework. To measure PBD-R’s improvements, rigorous, quantifiable metrics of physical fidelity are required. However, systematically measuring how accurately a simulator reproduces real-world physics remains an open problem. To that end, we introduce a physics benchmark consisting of seven tests with their corresponding analytical solutions. This benchmark is hardware-free, solver-agnostic, and easily extensible, enabling systematic comparison of simulation methods. Moreover, it complements common evaluation practices that focus on task-level metrics such as reward curves or sim-to-real transfer [ 3 , 10 , 1 ] and enables direct observation of error by measuring deviation from analytical solutions. Using this benchmark, we conduct an extensive evaluation of PBD-R, comparing it against PBD and MuJoCo’s [ 24 ] rigid-body solver. Our results show that PBD-R consistently outperforms standard PBD and achieves competitive accuracy to MuJoCo, while requiring fewer computational resources. In summary, our contributions are: (1) A revised PBD formulation, PBD-R, with a novel constraint and velocity update modification that achieves a significantly improved physical accuracy. (2) A solver-agnostic physics benchmark comprising seven tests with analytical reference solutions, providing a general framework for quantitative evaluation of physical accuracy in physics simulators. (3) An extensive empirical evaluation comparing PBD-R with standard PBD and MuJoCo across all seven benchmark tests. Collectively, our work paves the way toward a single high-fidelity, fully differentiable, and computationally efficient simulator that accurately captures the complexity of real-world physics. II Related Work Modern robotics simulators rely on several physics subsystems to model objects with different material behaviors. For example, MuJoCo’s [ 24 ] core engine simulates articulated rigid bodies with a soft‑constraint, convex‑optimization contact solver. However, deformable objects such as cables, cloth, and soft boxes are modeled as FEM‑style objects with their own contact and material models. PyBullet [ 7 ] similarly pairs a constraint-based multibody rigid solver with a separate soft-body module that implements cloth and volumetric deformables using FEM, mass–spring, or position-based dynamics backends, while experimental SPH fluid modules exist as yet another particle-based subsystem. NVIDIA Isaac [ 17 , 21 ] builds on PhysX to expose three distinct GPU-accelerated pipelines: a rigid and articulated body solver, a finite-element-based deformable body solver operating on tetrahedral meshes, and a position-based dynamics particle system for fluids, granular media, and cloth. Relying on fragmented simulation subsystems inherently complicates cross-material dynamics and breaks differentiability, resulting in brittle pipelines that fail to capture the contact-rich interactions required for real-world robotics. Particle-based simulators present a compelling alternative. A unified particle formulation enables diverse material behaviors to be modeled within a single framework [ 14 ] . Their mesh-less nature further provides practical advantages, including collision checking without convex decomposition [ 25 ] , faster collision computation [ 20 ] , and digital twin construction by mapping point clouds to simulation-ready particle sets [ 2 ] . Among particle-based methods, Position-Based Dynamics is a popular choice in computer graphics due to its computational efficiency, numerical stability, and ability to create visually plausible outputs. PBD supports a