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:20260817T171530Z LOCATION:Hall B7 (1)\, B Block\, Level 7 DTSTART;TZID=Asia/Tokyo:20241205T130000 DTEND;TZID=Asia/Tokyo:20241205T131400 UID:siggraphasia_SIGGRAPH Asia 2024_sess132_papers_1034@linklings.com SUMMARY:PDP: Physics-Based Character Animation via Diffusion Policy DESCRIPTION:Takara Truong, Michael Piseno, Zhaoming Xie, and Karen Liu (St anford University)\n\nGenerating diverse and realistic human motion that c an physically interact with an environment remains a challenging research area in character animation. Meanwhile, diffusion-based methods, as propos ed by the robotics community, have demonstrated the ability to capture hig hly diverse and multi-modal skills. However, naively training a diffusion policy often results in unstable motions for high-frequency, under-actuate d control tasks like bipedal locomotion due to rapidly accumulating compou nding errors, pushing the agent away from optimal training trajectories. The key idea lies in using RL policies not just for providing optimal traj ectories but for providing corrective actions in sub-optimal states which gives the policy a chance to correct for errors caused by environmental st imulus, model errors, or numerical errors in simulation. Our method, Physi cs-Based Character Animation via Diffusion Policy (PDP), combines reinforc ement learning (RL) and behavior cloning (BC) to create a robust diffusion policy for physics-based character animation. We demonstrate PDP on pertu rbation recovery, universal motion tracking, and physics-based text-to-mot ion synthesis.\n\nRegistration Category: Full Access, Full Access Supporte r\n\nLanguage Format: English Language\n\nSession Chair: Yi Zhou (Roblox)\ n\n URL:https://asia.siggraph.org/2024/program/?id=papers_1034&sess=sess132 END:VEVENT END:VCALENDAR