Chrono::Ray: A Distributed Framework for High-Throughput Simulation-Based Analysis of Multibody Systems

Khailanii Slaton, Dan Negrut · 2026

Read Chrono::Ray: A Distributed Framework for High-Throughput Simulation-Based Analysis of Multibody Systems by Khailanii Slaton, Dan Negrut (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-2605-13767.txt.

Introduction

Khailanii Slaton and Dan Negrut present Chrono::Ray, a shared frame for high-throughput test-based study of multi-part systems. Chrono::Ray joins the Chrono multi-part motion engine with the Ray shared compute platform. That gives a modular workflow with user-friendly shortcuts for large batches of test trials without direct servers handling. Uses are shown through setting finding for a multi-part lunar lander model and design of experiments for a soil ground-motion model. Chrono::Ray is open-source software in the Project Chrono family.

Catalog record

CREATORKhailanii Slaton, Dan Negrut
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25291
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.13767

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Chrono::Ray: A Distributed Framework for High-Throughput Simulation-Based Analysis of Multibody Systems Khailanii Slaton, Dan Negrut Abstract Large-scale simulation studies can provide invaluable insights across computational engineering efforts, but they are often computationally demanding, requiring the use of distributed computing, which is itself not a simple task. Chrono::Ray addresses this challenge by integrating the high-fidelity multibody dynamics simulation engine Chrono with the open-source distributed computing platform Ray. The result is a modular workflow framework providing user-friendly abstractions for large-scale engineering simulation studies, supporting scalable orchestration of large ensembles of simulation trials without requiring users to directly manage distributed infrastructure. The current capabilities of the framework are demonstrated through two representative examples: parameter recovery for a multibody lunar lander model, and design of experiments for parameters of a continuum terramechanics model. Chrono::Ray is a part of the larger Project Chrono ecosystem and is released as an open-source software package, with source code available at https://github.com/uwsbel/chrono-ray.git . 1 Motivation Computer simulation plays a crucial role in modern engineering design, existing as a powerful tool for system development, validation, and performance evaluation prior to physical prototyping. Specifically, high-fidelity simulation environments allow engineers to analyze the performance and behavior of complex systems under a wide range of operating conditions. However, simulation-based analysis is inherently affected by two complementary sources of uncertainty. Epistemic uncertainty arises from incomplete knowledge of model parameters or system behavior, possibly reduced by gathering additional information. In simulation, this manifests as situations where the analyst knows the model is imperfect but lacks sufficient data to fully constrain it. Aleatoric uncertainty is associated with irreducible stochastic variability in physical processes, and cannot be eliminated through additional measurement or modeling effort. In simulation, this can manifest as variability in initial conditions, manufacturing tolerances, or environmental disturbances. To address these uncertainties, robust system analysis must be employed. This typically involves many simulation trials in which model parameters are varied and the system is evaluated with respect to some common objective. For computationally intensive multibody models, these large-scale simulation efforts can quickly become prohibitively expensive when executed sequentially. As a result, scalable workflow orchestration and distributed execution have become increasingly important components of modern computational engineering and scientific computing pipelines. To address this computational bottleneck, we introduce Chrono::Ray, a distributed simulation framework consisting of the integration of the high-fidelity multibody dynamics engine Chrono and the Ray distributed computing platform. Rather than requiring users to directly manage distributed execution, resource scheduling, or worker coordination, Chrono::Ray is a package of several modular, workflow-oriented interfaces that abstracts the complexity of distributed execution and provides ready-made support for tasks such as parameter estimation, Bayesian inference, and robust design optimization. The resulting tool brings large-scale statistical simulation within reach for computationally intensive simulation studies where sequential execution is no longer practical. 2 Ray: A Distributed Computing Platform Ray [ 2 ] is an open-source distributed computing framework designed to unify parallel and distributed execution under a single, Python-native abstraction, allowing users to accelerate their workflows without expert knowledge of distributed computing. Originally developed at the University of California, Berkeley, it is currently maintained as an active open-source project 1 1 1 https://github.com/ray-project/ray that has been commonly utilized in both academic research and industrial machine learning efforts. 2.1 Core Architecture At its foundation, Ray provides a task-based execution model in which arbitrary Python functions (decorated with @ray.remote ) are dispatched to a distributed scheduler that manages resource allocation, fault tolerance, and data movement transparently. Users need not write explicit message-passing code or manage worker processes directly; the framework handles all of this under the hood. Ray supports execution on a single multicore workstation as well as on large heterogeneous clusters, making it straightforward to scale a workflow from a laptop to a cloud environment without modifying application code. Ray exposes two main levels of API: • Ray Core is the lower-level API that allows users to implement custom distributed workflows with more control over task parallelism, actor lifecycles, and shared object stores. • Ray AIR (AI Runtime) is a higher-level API that provides ready-made templates for common machine learning and optimization workflows. It is composed of several specialized libraries, including Ray Tune for hyperparameter search and Ray Train for distributed model training. Additionally, libraries in this layer integrates seamlessly with the broader Python scientific computing ecosystem (NumPy, SciPy, PyTorch, scikit-learn, and others). 2.2 Ray Tune and Search Algorithms Of the libraries that compose Ray AIR, Ray Tune [ 3 ] is the most directly relevant to the engineering simulation workflows targeted by this work. Originally designed for machine learning hyperparameter optimization, Ray Tune’s underlying execution model is general enough to treat any parameterized Python callable as a “trial,” making it equally well-suited to driving large ensembles of physics simulations. Its key components are as follows: • Search algorithms determine how for each trial, how values for parameters are selected. Available options include random search, grid search, Bayesian optimization (via BoTorch or Optuna), BOHB, and CMA-ES, among others. • Trial schedulers manage the prioritization and, where applicable, early termination of trials. For simulation workloads that report intermediate metrics at runtime checkpoints, schedulers such as ASHA and Hyperband can terminate unpromising trials early to conserve compute resources. • Parameter space distributions allow users to specify prior distributions over each parameter, including Uniform , LogUniform , RandInt , Randn , and arbitrary grid-valued choices. Because Ray Tune requires no modification to accept a Chrono simulation function in place of a neural network training loop, it serves as the natural backbone for the statistical workflows implemented in Chrono::Ray. 3 Chrono::Ray: Framework Description Chrono::Ray is a modular software interface that bridges PyChrono, the Python binding of the Project Chrono multibody dynamics engine [ 1 ] , with the Ray distributed computing platform. Its central design goal is to allow engineers and researchers to leverage high-throughput computing (HTC) for engineering simulation studies with minimal additional cognitive load. The main idea is that a user familiar with PyChrono only needs to allocate most of their effort to defining the simulation function and a parameter space. They then pass this information to their desired Chrono::Ray instance, in which the distributed execution, result aggregation, and interface to the selected analysis workflow is handled. 3.1 Design Philosophy The goal of Chrono::Ray is to provide user-friendly interfaces that handle most of the complexity of the desired analysis workflow, requiring the user to know little to nothing about distributed computing or Ray. This is achieved by wrapping the relevant Ray internals within the framework itself, so the user is never required to dig through Ray documentation directly. Extensive inline documentation and per-class info() functions are provided to guide the user through configuration and usage. Each module follows a similar design pattern, in which the user mainly interacts with a given workflow via the following two primitives: • simulate_fn(config) a user-defined PyChrono simulation script wrapped as a callable that accepts a config dictionary of sampled parameter values and returns a dictionary of floats to be used transparently by the workflow’s objective function. • param_sample_space : dict[str, ChR_Distr] a dictionary mapping parameter names to sampling distributions defined in the ChR_Distr (e.g., uniform , loguniform , choice ). The workflow’s search algorithm samples parameter values from these

Read the complete text