SIGGRAPH Asia 2026 · Workshop
Open GPU Physics, Differentiable Simulation, and Multi-Solver Ecosystems for Physical AI and Embodied Robotics
Overview
One path from programmable physics to research questions, solver choices, and policy evaluation.
An NVIDIA Warp demonstration sets up a shared contact-and-manipulation example. The Taiwan-led Beyond Rigid Bodies team then runs research clinics and group exercises on the four pillars below, turning them into modeling assumptions and validation requirements. Newton and Isaac Lab sessions close the day by testing those requirements against multi-solver interfaces and robot-learning workflows.
Participants jointly produce a task-to-solver decision record, an evaluation checklist, and prioritized open questions. Alternative engines, critical comparisons, and negative results are welcome. No GPU or internet connection is needed for the core activities.
Four pillars
FEM, MPM, IPC, frictional contact, compliant materials, cloth, and granular media expose behavior absent from a simple rigid-contact baseline. Participants identify the physical effect that changes task success, the cost of representing it, and a diagnostic scene that tests whether increased fidelity helps.
System identification, inverse design, morphology optimization, and policy learning require different derivatives and tolerances for approximation. Participants examine contact discontinuities, stiffness, and long-horizon sensitivity, then choose a gradient check or alternative optimization strategy.
Reconstruction, neural scene representations, video-based system identification, and policy evaluation connect observations to a usable physical model. Participants list identifiable parameters, unresolved ambiguities, and held-out observations needed to validate a twin.
Learned dynamics, neural constitutive models, and structured predictive models can complement physical solvers. Participants ask which predictions a controller needs, how to test consistency over time, and what failures reveal limits to generalization.
Research & industry
9 speakers
Professor · The University of Hong Kong
Researches character animation, physics-based simulation, and contact-rich control for robot learning.
Assistant Professor · Tsinghua University (IIIS)
Develops differentiable simulation and computational design methods for soft robots, actuated mechanisms, and inverse design.
Assistant Professor · Columbia University (RoboPIL)
Studies structured world models, deformable-object manipulation, and neural dynamics for closed-loop robot control.
Assistant Professor · Brown University (IVL)
Studies 3D and 4D scene understanding, hand–object interaction, and reconstruction for robotics and digital twins.
NVIDIA
Connects physical AI, robotics, and simulation workflows, coordinating the NVIDIA sessions and their handoff from the BRB research clinics.
Senior Staff Developer Advocate · NVIDIA
Builds robotics and simulation demonstrations with Newton, NVIDIA Warp, and Jetson, and supports practical workshop exercises.
Physical AI Developer Relations · NVIDIA
Supports NVIDIA robotics and simulation communities through Isaac Sim, Isaac Lab, Newton, and NVIDIA Warp.
Senior Product Manager · NVIDIA
Works on high-performance simulation and real-time physics through NVIDIA Warp and Newton.
Robotics Software Product · NVIDIA
Leads robotics software product for simulation, robot learning, and robot autonomy.
Schedule · Updated November 5, 2026
Kuala Lumpur time (UTC+8). Session slots may still change.
Organizing teams · Introduce the shared manipulation brief and record questions to revisit at the close.
NVIDIA team · Predict a parameter change, inspect the result, and document the baseline’s limits.
Informal questions and case collection.
Srinath Sridhar · Identify observations, unknown parameters, and a calibration check.
Yunzhu Li · Define a useful prediction and a test of generalization.
Coffee.
BRB team · Participants pitch cases and form working groups across the four themes.
Tao Du · Examine gradient assumptions and specify a check or alternative.
Informal exchange.
Taku Komura · Identify a missing physical effect and a diagnostic scene.
NVIDIA team · Map requirements, explore a guided example, and critique solver choices to produce a decision record.
Transition between demonstrations.
NVIDIA team · Walk through a task, examine a policy rollout and perturbation, and develop an evaluation checklist.
Organizing teams and participants · Revisit initial questions, finalize checks and open problems, and identify volunteer follow-up owners.
The four 30-minute BRB research clinics (10:15–11:15, 11:45–12:15, and 13:30–14:00) each combine 15 minutes of technical input, 10 minutes of participant work or discussion, and 5 minutes of synthesis.
Workshop team
10 organizers and contributors
NTU · MoonShine Animation Studio
Master's student and technical practitioner researching neural PDEs, differentiable simulation, and robotics.
Research Manager · National University of Singapore
Works on real-time hyperelastic simulation, frictional contact, and simulation infrastructure for robotics.
Researcher · National University of Singapore
Studies physical simulation for policy learning, safety-aware control, and learned models for robotic systems.
Assistant Professor · University of British Columbia
Leads the UBC PhysAI Lab, studying computational design, fabrication, control, and physical simulation.
Independent Researcher
Develops robust solvers for deformable bodies and physics-based character animation.
Associate Professor · National Yang Ming Chiao Tung University
Studies computer graphics, 3D vision, neural rendering, and neural scene representations for embodied agents.
Professor · National Taiwan University
Leads NTU's Computer Graphics Group, with research in computer graphics, interactive techniques, and multimedia systems.
NVIDIA
Connects physical AI, robotics, and simulation workflows, coordinating the NVIDIA sessions and their handoff from the BRB research clinics.
Senior Staff Developer Advocate · NVIDIA
Builds robotics and simulation demonstrations with Newton, NVIDIA Warp, and Jetson, and supports practical workshop exercises.
Physical AI Developer Relations · NVIDIA
Supports NVIDIA robotics and simulation communities through Isaac Sim, Isaac Lab, Newton, and NVIDIA Warp.
Join us at SIGGRAPH Asia 2026 in Kuala Lumpur.