Source-linked AI summary
Demo: Closed-Loop Sionna-Isaac Sim Co-Simulation Framework for Wireless-Aware Robot Navigation over ROS 2
Yi Shen Lim, Seungeun Oh, Jihong Park
TL;DR
Network-controlled robots carry their receivers, but no single simulator combines robotic physics with site-specific channel modeling, and prior couplings are offline or stochastic. This work couples Isaac Sim and Sionna over ROS 2 and demonstrates a campus-scale navigation application quantifying the connectivity–travel-time trade-off.
Problem
Robotic simulators lack channel models and wireless simulators lack physics, while existing couplings use statistical channels, offline ray tracing, or stochastic propagation.
Source: §1 · p. 1Method
The framework couples Isaac Sim and Sionna over ROS 2, using a shared OSM-derived twin and publishing ray-traced coverage as a costmap layer for planners.
Source: §1 · pp. 1–2Results
A campus-scale navigation demo quantifies the connectivity–travel-time trade-off produced by the framework.
Source: §1 · p. 1Takeaways & Limitations
The ray-traced coverage map is a standard costmap layer that any ROS 2 planner can consume.
Source: §1 · p. 1The evaluated scene uses deterministic channels, ground-truth poses, reciprocal channels, a shared map, and best-server association without inter-cell interference.
Source: §2 · p. 2
Abstract
from arXiv · showhide
A robot that offloads its control loop to the network carries the receiver with it, so link quality is decided by where it goes. Simulating this requires both a physics engine and a site-specific propagation model at once; to our knowledge no simulator natively unifies both, with existing couplings of the two limited to offline analyses. We demonstrate a real-time co-simulation framework coupling NVIDIA Isaac Sim and NVIDIA Sionna over ROS 2 that closes the perception-action-communication (PAC) loop between them. Sionna ray-traces the base-station-to-robot channel over the exact geometry Isaac Sim simulates on, rather than modeling it stochastically, and feeds channel states back into the control loop in real time. Ray-traced on GPU, the coverage map is refreshed in ~16ms (~60 Hz), fast enough for real-time control. To showcase the framework's utility, we implement a wireless-aware navigation application in an OpenStreetMap(OSM)-derived SUTD campus twin with two Nova Carter robots: the closed-loop planner eliminates communication outage at only +7.4% traversal time over the shortest-path baseline (which spends 7.9 s of its 81.2 s run disconnected).
1 Introduction
Network-controlled robots carry their receivers, so motion and site-specific propagation must be simulated together. The framework couples Isaac Sim and Sionna over ROS 2 to bring ray-traced channel states into the perception-action-communication loop.
1 Introduction
- Cloud robotics, teleoperation, and autonomous logistics rely on low-latency wireless links, whose quality varies with robot position as structures reflect, diffract, or block signals.The robot’s path therefore determines its connectivity.Source: §1 · p. 1
- No single simulator combines robotic physics with site-specific channel modeling and online channel feedback to the motion planner.Prior approaches use statistical channel maps, offline ray tracing, stochastic propagation, or mobility that is given rather than planned.Source: §1 · p. 1
- The framework couples Isaac Sim, Sionna, and ROS 2, with the ray-traced coverage map published as a costmap layer for ROS 2 planners.The campus-scale demo quantifies the resulting connectivity–travel-time trade-off.Source: §1 · p. 1
- Both simulators must share scene geometry so ray-traced cells align with navigation cells and moved obstacles cast physical and RF shadows.The physics, ray-tracing, and planning loops run at different timescales and must interlock without stalling.Source: §1 · p. 1
- An OSM-derived twin supplies the physics geometry, ray-tracing scene, and 0.5 m navigation grid; asynchronous, message-triggered ROS 2 nodes coordinate the loops.The shared grid makes ray-traced cells register with navigation cells, while moved obstacles update all three representations.Source: §1 · p. 2
2 System Architecture
The architecture uses a shared campus twin for robot physics, ray-traced coverage, and navigation, then feeds wireless and obstacle costs to a planner through ROS 2.
2 System Architecture
- Two Nova Carter robots traverse a 540×540 m OSM-derived SUTD campus twin at 3.0 m/s, with poses published over ROS 2 at 60 Hz.The 0.5 m navigation grid derives from the same twin, and onboard sensing adds runtime obstacles.Source: §2 · p. 2
- Sionna ray-traces RSS from two 5G base stations over the shared geometry on a 1.5 m grid, refreshing the map in ∼16 ms (∼60 Hz).The map is published as a ROS 2 OccupancyGrid; retracing is triggered by transmitter motion or geometry changes.Source: §2 · p. 2
- Each cell’s wireless cost rises below RSSmin=−80 dBm, with Δ=30 dB spanning Nav2’s 0–254 cost range.RSSmin is the lowest power at which the link budget sustains the command stream with margin.Source: §2 · p. 2
- The planner searches Ctotal(x,y) = Cobs(x,y) + α·Cw(x,y) with A*, where α=0 gives shortest-path planning and larger α weights detours around dead zones.A new search is triggered when the costmap changes, and velocity commands return to Isaac Sim over ROS 2.Source: §2 · p. 2
3 Demonstration
The demonstration evaluates a wireless-aware planner against a shortest-path baseline and shows the connectivity–travel-time trade-off. It also presents live waypoint setting over the RSS heatmap.
Scenario 1: Interactive Closed-Loop Operation
- Attendees set waypoints in the operator view over the RSS heatmap during interactive closed-loop operation.Source: §3 · p. 2
Scenario 2: System Performance & Closed-Loop Evaluation
- 7.9 s of outage over an 81.2 s run for α=0 falls to 0 s at α=10−1, with traversal time +7.4%.The α=10−1 traversal takes 87.0 s and has a minimum RSS of −60.9 dBm.Source: Figure 2 · p. 2
- 7.3 ms of steady-state processing updates costmaps, while the full loop takes 4.73±0.10 s, dominated by global A* search.The reported 7.3 ms includes state ingestion, ray tracing, and costmap fusion; initial ray-tracing JIT takes 118 ms.Source: Figure 2 · p. 2