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:20260817T171536Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T113100
DTEND;TZID=Asia/Tokyo:20241204T114300
UID:siggraphasia_SIGGRAPH Asia 2024_sess114_papers_382@linklings.com
SUMMARY:Autonomous Character-Scene Interaction Synthesis from Text Instruc
 tion
DESCRIPTION:Nan Jiang (Peking University, Beijing Institute for General Ar
 tificial Intelligence); Zimo He (Peking University); Zi Wang (Beijing Univ
 ersity of Posts and Telecommunications); Hongjie Li (Peking University); Y
 ixin Chen and Siyuan Huang (Beijing Institute for General Artificial Intel
 ligence); and Yixin Zhu (Peking University)\n\nSynthesizing human motions 
 in 3D environments, particularly those with complex activities such as loc
 omotion, hand-reaching, and human-object interaction, presents substantial
  demands for user-defined waypoints and stage transitions. These requireme
 nts pose challenges for current models, leading to a notable gap in automa
 ting the animation of characters from simple human inputs. This paper addr
 esses this challenge by introducing a comprehensive framework for synthesi
 zing multi-stage scene-aware interaction motions directly from a single te
 xt instruction and goal location. Our approach employs an auto-regressive 
 diffusion model to synthesize the next motion segment, along with an auton
 omous scheduler predicting the transition for each action stage. To ensure
  that the synthesized motions are seamlessly integrated within the environ
 ment, we propose a scene representation that considers the local perceptio
 n both at the start and the goal location. We further enhance the coherenc
 e of the generated motion by integrating frame embeddings with language in
 put. Additionally, to support model training, we present a comprehensive m
 otion-captured dataset comprising 16 hours of motion sequences in 120 indo
 or scenes covering 40 types of motions, each annotated with precise langua
 ge descriptions. Experimental results demonstrate the efficacy of our meth
 od in generating high-quality, multi-stage motions closely aligned with en
 vironmental and textual conditions.\n\nRegistration Category: Full Access,
  Full Access Supporter\n\nLanguage Format: English Language\n\nSession Cha
 ir: Kai Wang (Simon Fraser University, ShanghaiTech University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_382&sess=sess114
END:VEVENT
END:VCALENDAR