Physics-Based Simulation
Overview
Physics-based simulation uses mathematical laws and numerical solvers to predict how materials, fluids, devices, robots, and environments change over time, supporting design, safety testing, and control when experiments are costly or risky. Traditional methods include finite-element and finite-difference solvers, material point and particle methods, and rigid-body contact engines; newer work combines them with differentiable simulation, physics-informed neural networks and operators, graph networks, and generative or foundation models. These methods serve multiphysics engineering, molecular and detector modeling, robotics, deformable graphics, and synthetic sensor data, often using GPU parallelism and sim-to-real calibration. The central tension is achieving physical fidelity and stable long-horizon behavior while reducing compute and labeled data, with evaluation increasingly checking real-world agreement, conservation, contact, and whether generated code solves the intended equations.
Latest papers
231 papers28 papers in the last four weeks, up 8% on the four weeks before. 0.2% of all new papers.