BUMP: Body-driven Unified Multi-object Physics for HOI Synthesis

Key points

• Generating human-object interactions (HOIs) with multiple objects is challenging because the resulting motions arise from different underlying causes.1
• We then obtain responder object motions through physics simulation, allowing the framework to handle varying numbers and configurations of surrounding objects without expanding the learned trajectory output space.1
• To address the scarcity of paired full-body multi-object HOI data, we combine heterogeneous supervision from human-only motion, full-body HOI, and hand-centric manipulation data augmented with plausible body motion.1
• We further introduce outcome-conditioned interaction editing, which infers a local modification of the causal driver object trajectory from a desired responder object behavior without requiring an explicit target driver object trajectory.1

Sources (1)

• [1]BUMP: Body-driven Unified Multi-object Physics for HOI Synthesis
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 06:00 PM

  We present BUMP, a text-driven framework for multi-object HOI generation based on a simple principle: learn intentional motion and simulate physical consequences.

  Generating human-object interactions (HOIs) with multiple objects is challenging because the resulting motions arise from different underlying causes.

Extractive summary: sentences quoted from the sources.