GPU simulation enables large-scale 5G network and robot training together
Network-in-the-Loop at Scale: GPU-Batched 5G Simulation for Massively Parallel Robot Learning
Networking and Internet ArchitectureArtificial IntelligenceDistributed, Parallel, and Cluster ComputingRobotics
Summary
Training many robots at once requires simulating how they communicate over 5G networks, but existing methods either simplify the network too much or are too slow for lots of robots. The authors built Isaac-Net, a tool that simulates detailed 5G communication alongside thousands of robot environments all at once using GPUs. This helps capture realistic network delays and data freshness, improving how robot teams learn to work together with real communication constraints. Their experiments show Isaac-Net scales to about a million robots on one GPU while matching results of slower simulators.
What this means in practice
- •For multi-robot system developers: Simulate large fleets of robots communicating over 5G networks with realistic delays to improve training of coordinated behaviors.
- •For edge network engineers: Test and evaluate 5G network performance at scale for devices transmitting uplink data with realistic timing, all on a single GPU.
Authors
Zifan Zhang, Mingzhe Han, Kannan Athreya, Yuchen Liu
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