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:20250110T023312Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T154300
DTEND;TZID=Asia/Tokyo:20241204T155400
UID:siggraphasia_SIGGRAPH Asia 2024_sess120_papers_664@linklings.com
SUMMARY:MaskedMimic: Unified Physics-Based Character Control Through Maske
 d Motion Inpainting
DESCRIPTION:Technical Papers\n\nChen Tessler (NVIDIA Research), Yunrong Gu
 o (NVIDIA), Ofir Nabati and Gal Chechik (NVIDIA Research), and Xue Bin Pen
 g (NVIDIA)\n\nCrafting a single, versatile physics-based controller that c
 an breathe life into interactive characters across a wide spectrum of scen
 arios represents an exciting frontier in character animation. An ideal con
 troller should support diverse control modalities, such as sparse target k
 eyframes, text instructions, and scene information. While previous works h
 ave proposed physically simulated, scene-aware control models, these syste
 ms have predominantly focused on developing controllers that each speciali
 zes in a narrow set of tasks and control modalities. This work presents Ma
 skedMimic, a novel approach that formulates physics-based character contro
 l as a general motion inpainting problem. Our key insight is to train a si
 ngle unified model to synthesize motions from partial (masked) motion desc
 riptions, such as masked keyframes, objects, text descriptions, or any com
 bination thereof. This is achieved by leveraging motion tracking data and 
 designing a scalable training method that can effectively utilize diverse 
 motion descriptions to produce coherent animations. Through this process, 
 our approach learns a physics-based controller that provides an intuitive 
 control interface without requiring tedious reward engineering for all beh
 aviors of interest. The resulting controller supports a wide range of cont
 rol modalities and enables seamless transitions between disparate tasks. B
 y unifying character control through motion inpainting, MaskedMimic create
 s versatile virtual characters. These characters can dynamically adapt to 
 complex scenes and compose diverse motions on demand, enabling more intera
 ctive and immersive experiences.\n\nRegistration Category: Full Access, Fu
 ll Access Supporter\n\nLanguage Format: English Language\n\nSession Chair:
  Hao (Richard) Zhang (Simon Fraser University, Amazon)
URL:https://asia.siggraph.org/2024/program/?id=papers_664&sess=sess120
END:VEVENT
END:VCALENDAR