FIRE-3DV: Framework-Independent Rendering Engine for 3D Graphics using Vulkan

TL;DR

The paper presents FIRE-3DV, a framework-independent Vulkan-based rendering engine designed to modernize dynamic robotic simulators such as AMBF by overcoming legacy OpenGL limitations. Implemented as a standalone renderer with a lightweight AFI plugin, FIRE-3DV delivers high-fidelity visuals through Physically Based Rendering, multi-sample anti-aliasing, and ray-traced shadows while maintaining real-time performance, achieving sub-2 ms GPU times on scenes with millions of triangles. Quantitative results show substantial speedups over the legacy AMBF renderer (roughly 6–8×) and qualitative improvements in reflection accuracy, shadow precision, and edge antialiasing. The work demonstrates Vulkan’s viability for robotics simulation and outlines a clear path for broader adoption, including deeper ray-tracing pipelines and enhanced tooling for user workflows.

Abstract

Interactive dynamic simulators are an accelerator for developing novel robotic control algorithms and complex systems involving humans and robots. In user training and synthetic data generation applications, high-fidelity visualizations from the simulation are essential. Yet, robotic simulators often limit their rendering algorithms to preserve real-time interaction with the simulation. Advancements in Graphics Processing Units (GPU) enable improved visualization without compromising performance. However, these advancements cannot be fully leveraged in simulation frameworks that use legacy graphics application programming interfaces (API) to interface with the GPU. This paper presents a performance-focused and lightweight rendering engine supporting the modern Vulkan graphics API that can be easily integrated with other simulation frameworks to enhance visualizations. To illustrate the proposed method, our engine is used to modernize the legacy rendering pipeline of the Asynchronous Multi-Body Framework (AMBF), a dynamic simulation framework used extensively for interactive robotics simulation development. This new rendering engine implements graphical features such as physically based rendering (PBR), anti-aliasing, and ray-traced shadows, significantly improving the image fidelity of AMBF. Computational experiments show that the engine can render a simulated scene with over seven million triangles while maintaining GPU computation times within two milliseconds.

Figures (8)

• Figure 1: Overview of the proposed dynamic simulation framework. The simulation engine handles user input through interface devices and runs physics simulations on the scene, recording scene updates in shared memory. FIRE-3DV reads the scene updates from shared memory and renders the result with high visual fidelity.
• Figure 2: Overview of FIRE-3DV's render loop from the scene update to the presentation of the image to the screen. The scene geometry is rendered in the Main Pass, and its resulting image is used as a sampled texture by the Post-Processing Pass. The ImGui pass renders the GUI overtop the image and the result is presented to the screen.
• Figure 3: Material examples using the PBR reflection model. For (a)-(c), normal incidence Fresnel reflectance measurements are used to create physically based metallic materials. For (d)-(e), the roughness parameter allows for distinction between glossy and matte materials.
• Figure 4: Comparison between anti-aliasing disabled (left) and enabled (right). Aliasing artifacts are due to undersampling of the scene in high-contrast areas. The top images show an improvement in edge jaggedness. The bottom images show an improvement in specular highlight clarity.
• Figure 5: Demonstration of the engine's ray-traced omnidirectional shadows in a scene with two light sources. A ray query traverses the acceleration structure from the fragment shader to determine the light visibility of the sampled point in the scene.