NEXUS: Neural Energy Fields for Physically Consistent Contact-Rich 3D Object Dynamics

NEXUS is a neural energy-field framework designed for modeling contact-rich 3D object dynamics, representing objects as structural graphs and employing dynamic contact graphs. It utilizes a Hamiltonian approach to formulate motion through energy and dissipation terms, enhancing long-horizon accuracy in trajectory predictions compared to existing learned and physics-structured methods. This advancement is significant for practitioners as it improves the physical plausibility of generated videos while maintaining competitive visual quality, enabling more realistic simulations in complex environments.