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:20241204T144500
DTEND;TZID=Asia/Tokyo:20241204T145600
UID:siggraphasia_SIGGRAPH Asia 2024_sess120_papers_586@linklings.com
SUMMARY:Robot Motion Diffusion Model: Motion Generation for Robotic Charac
 ters
DESCRIPTION:Technical Papers\n\nAgon Serifi (ETH Zürich, Disney Research);
  Ruben Grandia and Espen Knoop (Disney Research); Markus Gross (ETH Zürich
 , Disney Research); and Moritz Bächer (Disney Research)\n\nRecent advancem
 ents in generative motion models have achieved remarkable results, enablin
 g the synthesis of lifelike human motions from textual descriptions. These
  kinematic approaches, while visually appealing, often produce motions tha
 t fail to adhere to physical constraints, resulting in artifacts that impe
 de real-world deployment. To address this issue, we introduce a novel meth
 od that integrates kinematic generative models with physics-based characte
 r control. Our approach begins by training a reward surrogate to predict t
 he performance of the downstream non-differentiable control task, offering
  an efficient and differentiable loss function. This reward model is then 
 employed to fine-tune a baseline generative model, ensuring that the gener
 ated motions are not only diverse but also physically plausible for real-w
 orld scenarios. The outcome of our processing is the Robot Motion Diffusio
 n Model (RobotMDM), a text-conditioned kinematic diffusion model that inte
 rfaces with a reinforcement learning-based tracking controller. We demonst
 rate the effectiveness of this method on a challenging humanoid robot, con
 firming its practical utility and robustness in dynamic environments.\n\nR
 egistration Category: Full Access, Full Access Supporter\n\nLanguage Forma
 t: English Language\n\nSession Chair: Hao (Richard) Zhang (Simon Fraser Un
 iversity, Amazon)
URL:https://asia.siggraph.org/2024/program/?id=papers_586&sess=sess120
END:VEVENT
END:VCALENDAR