Awesome Neural Physics

Awesome Neural Physics.

Derived from the Awesome Neural Physics list, this interactive index explores the seamless fusion of neural models and physics simulation. We track the deep convergence of these fields: from injecting neural capabilities into classical solvers, to embedding physical simulators directly within neural architectures.

At its core, we focus on how neural models internalize physical laws—a crucial step toward building robust World Models. By curating cutting-edge research across fluids, cloth, and soft/rigid bodies, we hope this resource empowers your next breakthrough.

Category GuideOpen

Differentiable SimulationPipelines whose forward pass is an explicit physical simulator, with gradients propagated through that simulator.81

Neural RepresentationPhysics-aware neural state, field, or latent representations used to model physical processes.32

Neural SolverWorks where the solve, update, or simulation computation is replaced or driven by a learned solver.70

ReconstructionRecovering geometry, motion, state, or parameters from observations.63

ControlControl-centered methods such as policy design, MPC, or action optimization for physical systems.29

OptimizationTarget-driven optimization of shape, topology, parameters, trajectories, or initial states.8

Embodied AIValidation in robotic or embodied settings such as manipulation, locomotion, or embodied interaction.51

EngineClearly reusable engines, frameworks, or toolkits rather than one-off task-specific methods.15

FractureFracture, crack propagation, or material failure is a central simulation phenomenon.2

ContactContact interaction is one of the main technical axes of the paper.26

SoundAcoustics or sound is a primary simulation phenomenon.2

AvatarHuman-avatar-centric settings such as clothed humans or body-driven garment dynamics.18

SurveySurvey or review papers.5