Robotics: Science and Systems (RSS) 2026
NeuralActuator: Neural Actuation Modeling for Robot Dynamics and External Force Perception
Learned actuator dynamics for robot simulation and sensorless force perception.
PhysAI Lab
Research
We study robot manipulation and physical world modeling through simulation and machine learning.
01 / Research direction
Robotics and world models
We connect robot interaction, actuation modeling, and physical world construction to simulation and robot learning.
arXiv preprint · 2026
Agentic Real2Sim: Physics-based World Modeling with Vision-Language Agents
Recovering physics-based worlds from real interaction using vision-language agents.
Preprint, arXiv:2609.19600 · 2026
WorldContact: A Contact-Centric World Model for Scalable Robot Learning
Contact-centric deformable world modeling for robot learning.
Preprint, arXiv:2609.33208 · 2026
WorldAgent: Verification-Guided Agentic Physical World Construction
A verification-guided agent that constructs and checks physically simulatable worlds.
02 / Research direction
Neural simulation
We combine numerical methods and neural representations to simulate physical dynamics efficiently, across materials, geometries, and discretizations.
SIGGRAPH Asia 2026 · Conference Track
WorldParticle: Unified World Simulation of Lagrangian Particle Dynamics via Transformer
One Transformer architecture spanning fluids, cloth, solids, granular materials, and molecular dynamics.
ACM SIGGRAPH/Eurographics Symposium on Computer Animation (SCA), Computer Graphics Forum · 2026
Closing Trajectories: Equation-Free Cyclic Animation via Koopman Surrogates
Equation-free cyclic animation through learned Koopman surrogates.
ACM Transactions on Graphics (Proceedings of SIGGRAPH) · 2025
Fast Subspace Fluid Simulation with a Temporally-Aware Basis
A temporally-aware reduced basis for fast fluid simulation.
International Conference on Learning Representations (ICLR) 2023 notable-top-25%
CROM: Continuous Reduced-Order Modeling of PDEs Using Implicit Neural Representations
03 / Research direction
Geometry and scientific computing
We develop geometric representations and optimization tools for physical constraints, computational design, and scientific computing.
SIGGRAPH 2025
Shape Space Spectra
Eigenanalysis across continuously parameterized shape families for simulation and design.
SIGGRAPH 2025
Lifting the Winding Number: Precise Discontinuities in Neural Fields for Physics Simulation
Neural fields that represent precise cuts and discontinuities for physics simulation.
International Conference on Learning Representations (ICLR) · 2026
Physically Valid Biomolecular Interaction Modeling with Gauss-Seidel Projection
Differentiable Gauss–Seidel projection for physically valid biomolecular structures.
Preprint, arXiv:2609.36277 · 2026
From Surfaces to Volumes: Registered Geometry for Protein Representation Learning
Registered volumetric geometry for protein representation learning.
A connected research community
Research community
Our work connects visual computing, mechanics, scientific machine learning, and robotics. Peter is affiliated with UBC’s AI4Science community and Institute of Applied Mathematics.
Peter and his collaborators also work with industry on simulation challenges. Learn more about these collaborations on Peter’s research profile ↗.