NIFTY, or Neural Interaction Fields for Trajectory Synthesis, is a framework developed by a group of researchers from the University of Michigan, Google, NVIDIA, and Stanford University. This system addresses the challenge of generating realistic 3D motions of humans interacting with objects in a scene, with a specific focus on the intricate details of the “last mile” of object interaction. To achieve this, the team created a neural interaction field attached to specific objects, guiding the sampling of an object-conditioned human motion diffusion model. To support interactions where data is scarcely available, an automated synthetic data pipeline is also proposed.
Publication date: 14 Jul 2023
Project Page: https://nileshkulkarni.github.io/nifty
Paper: https://arxiv.org/pdf/2307.07511.pdf
