Replicate What Varies: Resource Sharing for Domain-Randomized Embodied Simulation

Zixu Hao; Weijun Wang; Xiangyu Li; Liang Mi; Hao Wu; Yunxin Liu; Ting Cao

cs.DCOct 24, 2026

Replicate What Varies: Resource Sharing for Domain-Randomized Embodied Simulation

Authors: Zixu Hao, Weijun Wang, Xiangyu Li, Liang Mi, Hao Wu, Yunxin Liu, Ting Cao

Organizations: Institute for AI Industry Research (AIR), Tsinghua University · Nanjing University

Abstract

Simulation provides a low-cost sandbox where embodied agents can attempt tasks, observe outcomes, and learn from experience. Running many diverse simulation environments concurrently can consume substantial CPU and GPU memory because independently constructed scenes duplicate resources that remain unchanged across instances. We present Patchwork, a resource-sharing system for domain-randomized simulation, built around the representation $E=T+Δ$ : a shared template contains common resources, and a patch records the mutations needed to construct a target environment. Patchwork separates randomization sampling from replay and selects reusable templates together with patches that preserve each target environment's intended configuration. Its layered runtime shares compiled model structure and immutable assets while keeping mutable parameters and simulation state private to each instance. We implement Patchwork on robosuite and MuJoCo and evaluate it on LIBERO-plus and RoboCasa365 workloads. At 32 replicas, shared-template execution reduces CPU RSS by 88-92% relative to process-per-environment execution. Controlled experiments show how sharing shifts memory growth toward template diversity and unique asset content rather than instance count, within the supported sharing boundaries.

Figures & tables

Figure 1: Resource sharing across diverse simulation environments.

Sim. Env.	Framework	Physics / Rendering
LIBERO [ 18 ]	robosuite [ 41 ]	MuJoCo [ 31 ] / OpenGL
RoboCasa [ 22 ]	robosuite [ 41 ]	MuJoCo [ 31 ] / OpenGL
ManiSkill [ 6 , 28 ]	SAPIEN [ 34 ]	PhysX [ 23 ] / Vulkan
Isaac Lab [ 20 ]	Isaac Sim [ 5 ]	PhysX / RTX
BEHAVIOR-1K [ 14 ]	OmniGibson	PhysX / RTX

Table 1: Representative embodied simulation software stacks.

Panel of Figure 2 — Figure 2: MuJoCo separates the compiled model ( mjModel ) from per-instance runtime state ( mjData ).

Figure 3: Domain randomization examples from LIBERO-plus and RoboCasa365.

Figure 4: Top: compiled models (a) and per-variant changes (b) in LIBERO-plus. Bottom: model and asset memory in LIBERO and RoboCasa. RoboCasa memory bars show medians of three seeds with min–max markers.

Figure 5: Overview of Patchwork. The pipeline transforms environment specifications into mutation representations, serves them with shared templates and patches, and materializes them through layered resource-sharing mechanisms.

Figure 6: Sampling produces an explicit mutation sequence that can be saved and replayed without further randomness.

Target	Example	Destination	Owner
Structure	object addition	model arrays	template
Model field	light diffuse	DR buffer	instance
Runtime state	object pose	mjData	instance
Metadata	task language	Python env	instance

Table 2: Logical mutation targets and their materialization destinations.

Figure 7: Compiled structure is shared within template groups, while immutable textures are deduplicated across groups, separately on CPU and GPU.

Figure 8: Template serving groups profiles by their unchanged structure and asset identities, selecting one representative per class and deriving member patches.

Figure 9: Resource use and throughput across replica counts. Bars show medians over three seeds with min–max markers.

		Steps/s		GPU (GiB)		RSS (GiB)
Config	$N$	off	on	off	on	off	on
indep.	1	45.5	39.6	0.65	0.68	2.70	2.72
indep.	8	226.3	162.3	1.83	2.15	10.66	10.82
indep.	32	348.9	199.4	5.89	7.14	27.61	27.67
shared	1	46.6	41.5	0.56	0.58	2.59	2.58
shared	8	420.4	267.8	1.16	1.32	5.11	5.44

Table 3: Effect of enabling camera observations on RoboCasa365. Values are medians over three seeds.

Figure 10: LIBERO-plus memory use as template diversity (a, b) and distinct variants per class (c) increase. Crosses mark independent-model baselines.

Figure 11: RoboCasa365 memory response to unique texture bytes (a) and texture sharing across templates (b), with

N{=}32

Index quantity	Value
Variant records	16,170
Distinct models	5,855
Selected templates	3,042 ( $-48.0$ %)
Additional models served	2,813 (6 zero-write)
Patch writes per additional model	median 6, max 22

Table 4: Static accounting of the LIBERO-plus offline index.

Figure 12: Template reuse under uniform LIBERO-plus request sampling. Bars show medians of three trials, with individual trials marked by dots.

Figure 13: RSS growth during template and instance construction, with

K{=}8

and

N{=}32

Figure 14: Model and GPU texture sharing ablation on RoboCasa365, showing medians of three seeds.

Robotics Simulation

Robotics simulation builds virtual robots and environments to model, train, and evaluate autonomous systems before real-world deployment.

See all 322 papers

Building a Scalable, Reproducible, Evaluatable, and Closed-Loop Simulation Environment Foundation for Embodied Intelligence
Jun 26, 2026Junwu Xiong, Yongjian Guo, Mingxi Luo +17Robotics Simulation Synthetic Data Generation for Robotics
1AI Infra Team at JDT · 2Tsinghua University · 3Peking University +1
CUA-Sandbox: Efficient Environments for Computer-Use Agent Reinforcement Learning
Sep 26, 2026Xin Yan, Zhengbo Jiao, Jiaqi Liu +11Computer-Use Agents Reinforcement Learning
A*STAR · BNU · HKUST +4
EmbodiedGen V2: An Agentic, Simulation-Ready 3D World Engine for Embodied AI
Jul 8, 2026Xinjie Wang, Liu Liu, Taojun Ding +93D Asset Generation Robotics Simulation
Horizon Robotics · WuwenAI

Jun 26, 2026J/K move · Enter open · S save

Building a Scalable, Reproducible, Evaluatable, and Closed-Loop Simulation Environment Foundation for Embodied Intelligence

Junwu Xiong, Yongjian Guo, Mingxi Luo, Ning Qiao+16

1AI Infra Team at JDT · 2Tsinghua University · 3Peking University+1

Robotics Simulation Synthetic Data Generation for Robotics Embodied AI

Open paper AskPDF arXiv

This paper presents a cloud-native simulation infrastructure framework for embodied intelligence that supports large-scale training, standardized evaluation, and simulation-based data collection. The framework unifies simulation environment generation, task execution, trajectory collection, model evaluation, data management, and cloud services into a scalable and reproducible platform. To address the high cost, limited scalability, and poor reproducibility of real-world robotic data collection, the framework adopts cloud-native technologies including elastic resource scheduling, containerized simulation, unified data management, and service-oriented system design, enabling efficient large-scale simulation for multi-model and multi-task workloads. Built on a four-layer architecture, the framework provides standardized environment assets, automated task generation, trajectory collection, benchmark evaluation, and closed-loop data optimization. It further integrates representative systems including D-VLA, RL-VLA3, Sword, and Pre-VLA to support scalable simulation, dynamic scheduling, visual augmentation, and real-time data filtering. We argue that cloud-native simulation infrastructure provides a unified foundation for data generation, model training, standardized evaluation, and real-world deployment, and will play a key role in the future development of embodied intelligence.

Abstract

Robotics Simulation

Explore similar work

Building a Scalable, Reproducible, Evaluatable, and Closed-Loop Simulation Environment Foundation for Embodied Intelligence

CUA-Sandbox: Efficient Environments for Computer-Use Agent Reinforcement Learning

EmbodiedGen V2: An Agentic, Simulation-Ready 3D World Engine for Embodied AI