- The paper introduces a unified simulation platform that integrates robotics, ray tracing, and wireless modules for consistent 6G ISAC analysis.
- It details a ROS-centric framework that synchronizes sensor and channel data in real-time, validated by high beam prediction accuracy across varied conditions.
- The design’s reproducibility and dual pipelines for real-world and synthetic scenes enable scalable ISAC research and rigorous algorithm benchmarking.
Motivation and Context
Integrated Sensing and Communication (ISAC) is central in 6G wireless networks, demanding tight temporal and spatial alignment among wireless channel propagation, platform mobility, and diverse sensor streams. Prior simulation tools for ISAC focus on isolated facets—robotics simulators provide rich physical and sensor modeling without electromagnetic realism, while wireless simulation platforms lack dynamic sensing and vehicle realism. Existing attempts to bridge this gap often involve custom, non-reproducible workflows that encumber research progress and hinder systematic dataset generation.
SimART addresses these fragmentation issues by delivering a unified, open-source multimodal simulation platform for ISAC research. The architecture is built around a ROS-centric integration backbone, leveraging mature external robotics, ray tracing, and wireless evaluation engines. A shared simulation clock, standardized message passing, and unified coordinate frames enforce tight temporal and geometric consistency across all modalities and modules.
The core platform consists of four interoperable components:
- Physics and Sensing Module: Implements a flexible ROS-defined contract; simulators such as AirSim, Gazebo, or CARLA can be integrated, provided they export standard pose and sensor topics.
- Ray Tracing Module: Uses Sionna RT, operating on a simplified but geographically and materially consistent mesh to efficiently compute propagation paths. Supports online synchronization, offline replay, and grid scan modes.
- Link and System Module: Employs Sionna SYS to simulate PHY/MAC behavior using site-specific channel impulse responses, with configurable numerology and spatial processing.
- Channel Knowledge Map (CKM) Generator: Exhaustively interrogates the simulated space to aggregate dense, high-fidelity channel and link-level priors, essential for location-aware ISAC algorithm development.
A single synchronized rosbag file captures all relevant simulation session data—sensor streams, propagation channels, KPIs, and CKM layers—structurally ensuring full reproducibility and ease of downstream evaluation or learning-based applications.
Scene Construction and Data Consistency
SimART provides dual pipelines for scene construction:
- Real-World Map Adaptation: Automates geometric and electromagnetic scene reconstruction from OpenStreetMap data, with 3D model generation managed via OSM2World and electromagnetic material annotation handled in Blender. This enables highly realistic urban simulations.
- User-Defined Scenes: Allows controlled synthetic or hypothetical scenarios, which are crucial for ablation studies or simulating not-yet-built environments. An automated Blender script ensures geometry simplification and material consistency for ray tracing.
In both pipelines, spatial assets for sensing/physics and electromagnetic modeling are always aligned in a common coordinate frame established at scene creation, minimizing manual alignment errors.
Data Collection and Usage Workflow
All simulation modules operate as independent ROS nodes. The platform’s data organization ensures that every message from every module is timestamped with the shared simulation clock and referenced to the unified coordinate tree. This organization supports:
- Sensor-level data (RGB, LiDAR, IMU, GPS, etc.) natively compatible with standard ROS toolchains.
- Per-link wireless data (CIRs, SINR, BLER, optimal beam indices) available as custom messages for direct learning and evaluation.
- Fully reproducible, replayable session data by design, promoting experiment rigor and simplifying cross-domain data inspection.
Typical workflows involve specifying a single configuration file, looping playback of trajectory data, and automatic alignment of all sensor and channel data—with subsequent visualization and analysis straightforward within standard ROS ecosystems.
Empirical Case Study: Multimodal Beam Prediction
A focused study demonstrates SimART’s end-to-end ISAC capability through multimodal beam selection for a low-altitude UAV scenario. Key experimental features:
- An urban scene is procedurally generated and transferred into AirSim and Sionna RT, ensuring geometric and material correspondence.
- Various environmental conditions (sunny, rainy, nighttime) are simulated, impacting both visual sensor data and electromagnetic surfaces.
- Joint vision- (YOLOv8-based UAV detection) and GPS-based features are used to predict the optimal beam from a 64-beam DFT codebook.
- The multimodal dataset is used to train and evaluate a baseline selection model across all conditions.
Key empirical results (top-1/top-3/top-5 accuracy):
| Condition |
Top-1 |
Top-3 |
Top-5 |
| Sunny |
97.86% |
99.83% |
100.00% |
| Rainy |
95.39% |
99.84% |
100.00% |
| Nighttime |
97.92% |
99.94% |
100.00% |
The strong results, including minor performance drops under more challenging visual and electromagnetic perturbations, empirically validate the robustness and fidelity of multimodal dataset generation and reveal that correct beams remain highly ranked across domain shifts.
Theoretical and Practical Implications
SimART’s design establishes a reproducible, extensible framework for 6G ISAC simulation. The modular, ROS-driven integration enables:
- Scalable simulation and dataset generation for a variety of scenarios (aerial, ground, indoor, maritime) with realistic mobility and radio channels.
- Separation of concerns between sensing frontends and wireless backends, fostering rapid prototyping and cross-validation of new algorithms.
- Dense CKM generation with physically interpretable priors for location-based ISAC methods, benchmarking, and data-driven model validation.
On a theoretical level, the consistent spatial-temporal data enable research on end-to-end spatially aware ISAC algorithms and facilitate the benchmarking of learning-based beam management, SLAM, and sensor fusion under a breadth of environmental conditions.
Future Directions
SimART opens several avenues for further development and research, including:
- Extending CKM generation to full 3D voxel grids for aerial and drone-heavy ISAC applications.
- Integrating more advanced sensor types (multi-spectral, mmW radar, etc.) for hybrid sensing-communication research.
- Incorporating dynamic and heterogeneous mobility patterns, multi-agent simulation, and complex regulatory or adversarial wireless scenarios.
- Scaling up for distributed and federated learning settings, leveraging full session reproducibility for robust ML benchmarking.
Conclusion
SimART provides a rigorously integrated, open multimodal simulation platform, unifying robotics, ray-tracing, and wireless evaluation for ISAC research. By ensuring modularity, reproducibility, and spatial-temporal fidelity, it significantly lowers the barrier for reproducible, scalable ISAC experiments. The public code release and compatibility with standard ROS ecosystems further positions SimART as a sustainable foundation for next-generation 6G ISAC simulation, benchmarking, and algorithmic advances (2605.13309).
