Network-in-the-Loop at Scale: GPU-Batched 5G Simulation for Massively Parallel Robot Learning

2026-10-01

AI Summary

This study addresses the challenges of coordinating GPU-based physical simulation with 5G network coupling latency in large-scale parallel robot training, as well as the performance bottlenecks inherent in packet-level simulators. To this end, it proposes Isaac-Net, a pioneering GPU-batched 5G NR engine that deeply integrates 5G protocol stack modeling into the Isaac Lab physics engine. By accurately simulating slot scheduling and network dependencies, this approach overcomes the limitations of conventional independent latency models, enabling the synchronous advancement of uplink transmission and physical simulation across thousands of environments. Experimental results demonstrate that Isaac-Net reproduces the median latency of ns-3 benchmarks while correcting Age-of-Information (AoI) tail deviations. Furthermore, it supports high-fidelity, network-in-the-loop training for million-scale robots on a single GPU at 83% of the original simulation speed.

Abstract

Massively parallel GPU simulators train multi-robot policies in thousands of environments, and many fleets use private Fifth-Generation (5G) networks, where each robot's delay depends on its teammates' traffic. Network-in-the-loop training places a simulated 5G network inside this loop. However, GPU robot simulators reduce the network to an independent delay per message, while packet-level simulators run one scenario per CPU process and cannot keep pace with thousands of parallel environments. To bridge this gap, we present Isaac-Net, a GPU-batched 5G New Radio (NR) module that advances the uplink of thousands of environments in lockstep with Isaac Lab physics. Isaac-Net simulates every slot, the 0.5~ms interval in which the base station decides which robots transmit, for all environments at once. Extensive experiments confirm that its NR engine reproduces the median delay of ns-3 5G-LENA across loads, with a median delay 5--10\% low on an unseen carrier and 9\% high at 32 robots per environment in closed loop. The engine also reproduces the Age of Information (AoI), the age of each robot's newest delivered report, while an independent delay per message leaves the AoI tail about three times too light. In a configuration validated against 5G-LENA, Isaac-Net keeps the network in the loop for about one million robots on one GPU at 83\% of the Isaac Lab rate without the network, measured under a random policy. Isaac-Net is open source at https://github.com/ZzZTripleZzZ/isaac-net

Problem

• Network-in-the-Loop
• Massively Parallel Robot Learning
• 5G Simulation
• GPU-Batched Simulation
• Age of Information

Innovation

• Network-in-the-Loop
• GPU-batched simulation
• 5G New Radio
• Massively parallel robot learning
• Age of Information