RadarSimPy v15.4.0 Release

Introduction

The v15.4.0 update is a performance-and-capability release. Mesh-based radar simulations run substantially faster on both CPU and GPU, long-range FMCW/stretch radars are now easier to model correctly, and animated 3D models can drive target motion directly. A handful of accuracy fixes also land in the ray-tracing core.

What’s New?

🎬 Animate Targets from 3D Model Files

A new animation_kit module lets you drive a target’s motion straight from an animated glTF/GLB 3D model:

• Author Motion Once: Design a spinning rotor, a turning wheel, or a flight path in your favorite 3D tool, export it as glTF, and simulate it as-is — no need to hand-code trajectories.
• Accurate Motion: Position, speed, rotation, and rotation rate are all derived directly from the animation, so fast-spinning or fast-turning parts stay physically correct.
• Broad Compatibility: Works with the common glTF animation styles used by most 3D authoring tools.

🚀 Much Faster Mesh Simulation

Simulations using mesh-based (SBR) targets are significantly faster, with no change to the results:

• Measured Speedups: Up to 8.2x faster on GPU and 6.3x faster on CPU, compared to the previous release.
• Better GPU-Build Performance: Fixed a case where GPU-enabled builds ran their CPU path much slower than a CPU-only build — one scene went from 65 seconds down to under 1 second.
• Faster RCS Simulation: Large radar-cross-section scenes that previously ran slower than expected on smaller GPUs are now handled correctly, with speedups of up to 11x.

📡 Support for Long-Range Radar Simulation

A new receive “gate delay” option makes it possible to accurately simulate long-range FMCW/stretch radars, where earlier versions could produce misleading results for targets far from the radar. This also enables realistic modeling of the radar’s usable range window.

🎛️ Simpler, Consistent GPU/CPU Selection

All simulation functions (sim_radar, sim_rcs, sim_lidar) now share the same automatic device selection: they use the GPU when available and fall back to the CPU automatically otherwise, with no code changes required on machines without a GPU.

🎯 Accuracy Improvements

Several edge cases that could produce subtly incorrect results in mesh-based simulations have been fixed, including scenes with very small or overlapping geometry, and certain material/angle combinations. Most users will see no visible difference; a few specific scene types will now be measurably more accurate.

Changelog Summary

Added

• New animation_kit module for simulating targets driven by animated glTF/GLB models.
• Automatic "auto"/"gpu"/"cpu" device selection for sim_radar, sim_rcs, and sim_lidar, with automatic fallback when no GPU is available.
• Receiver gate_delay option for accurate long-range FMCW/stretch radar simulation, plus related properties describing the radar’s usable range window.
• Expanded benchmarking tools for tracking simulation performance over time.

Changed & Optimized

• Mesh (SBR) simulation performance substantially improved on both CPU and GPU.
• RCS simulation performance improved, especially on large scenes and smaller GPUs.
• GPU-enabled builds now make full use of multi-core CPUs when running in CPU mode.
• Updated simulation documentation.

Fixed

• Several ray-tracing accuracy edge cases involving small/overlapping geometry, viewing angle, and certain material properties.
• Corrected a documentation inaccuracy about internal mesh-loading behavior.