BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171532Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T111300
DTEND;TZID=Asia/Tokyo:20241205T112700
UID:siggraphasia_SIGGRAPH Asia 2024_sess128_papers_618@linklings.com
SUMMARY:High-Throughput Batch Rendering for Embodied AI
DESCRIPTION:Luc Guy Rosenzweig, Brennan Shacklett, Warren Xia, and Kayvon 
 Fatahalian (Stanford University)\n\nIn this paper we study the problem of 
 efficiently rendering images for embodied AI training workloads, where age
 nt training involves rendering millions to billions of independent frames,
  often at low-resolutions and with simple (or no) lighting and shading, th
 at serve as the agent's observations of the world. To enable high-throughp
 ut end-to-end training, we design, and provide a high-performance GPU impl
 ementation of, a frontend renderer interface that allows state-of-the-art 
 GPU-accelerated batch world simulators to efficiently communicate with hig
 h-performance rendering backends for generating agent observations. Using 
 this interface we architect and compare two high-performance renderers: on
 e based on the GPU hardware-accelerated graphics pipeline and a second bas
 ed on a GPU software implementation of ray tracing.To evaluate these rende
 rers and encourage further research by the graphics community in this area
 , we build a rendering benchmark for this underexplored regime and find th
 at the ray tracing based renderer outperforms the rasterization based solu
 tion across the benchmark on a datacenter class GPU, while also performing
  competitively in geometrically complex environments on a high-end consume
 r GPU. When tasked to render large batches of independent 128x128 images, 
 the raytracer can exceed 100,000 frames per second per GPU for simple scen
 es, and exceed 10,000 frames per second per GPU on geometrically complex s
 cenes from the HSSD dataset.\n\nRegistration Category: Full Access, Full A
 ccess Supporter\n\nLanguage Format: English Language\n\nSession Chair: Man
 olis Savva (Simon Fraser University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_618&sess=sess128
END:VEVENT
END:VCALENDAR