UniPhy: Learning a Unified Constitutive Model for Inverse Physics Simulation

Figure 1 . We present UniPhy, a unified latent-conditioned neural model which learns a common latent space to encode the properties of diverse materials. At inference, given motion observations for a system with unknown material parameters, UniPhy allows material inference via differentiable simulation-based latent optimization. These inferred material latents can be used to simulate new trajectories that reflect the behavior of the underlying material. For example, in the first row, given the initial geometry of a toy and an optimized latent representing newtonian fluid (blue block), the geometry spreads when it hits the floor, whereas for an optimized latent representing elastic materials (purple block), it squeezes.