Emerging augmented and virtual reality (AR/VR) head-mounted displays (HMD) are critical for mission execution and training/simulation. HMDs require large, high-density, near-eye micro-displays. The rendering burden imposed by comfortable viewing update rates requires a significant amount of graphics computation. The size, weight, power, and bandwidth cost of rendering VR/AR content hinders the application and usage of HMDs. The introduction of foveated rendering with an eye/gaze tracking system will generate high-resolution content in the viewerÂ’s direct line of sight and low-resolution content in the periphery where the human visual system is less susceptible to detail. However, foveated rendering is still computationally expensive as it requires multiple rendering passes of the scene with varying levels of resolution. This stems from the fact that modern GPUs are architected to render a single view per update of a display. Re-architecting the traditional graphics pipeline to render foveated views in parallel can greatly reduce the SWaP cost of computation for HMDs. Building upon previous multi-view rendering research for extreme multi-view light-field displays, FoVI3D will develop an optimized foveated render pipeline than can be implemented in an untethered GPU like device that resides in the HMD to reduce the power, heat, and bandwidth requirements of foveated rendering.
