Embedding-Conditioned Gaussian Mixture for Online Physical Parameter Estimation
Estimate latent physical interaction parameters from pre-contact vision by replacing fixed semantic classes with a Gaussian-mixture posterior whose components are drawn from past, interaction-grounded parameter estimates. Continuous visual embeddings condition mixture weighting to retrieve and emphasize relevant prior interactions for the current scene region, while a support-aware fusion with a conditional generative prior handles cases with sparse interaction history. The method can incorporate learned embedding refinement to improve retrieval relevance without discretizing the representation.
Discovered 2026-11-10 · Sources checked 2026-11-10
Why it matters
Vision can be ambiguous about physical properties; semantic-class approaches collapse distinct physical regimes into a single label or aggregate. A mixture posterior conditioned on embeddings preserves multimodal physical beliefs and filters out unrelated past interactions, supporting more robust planning/control when the physical regime is uncertain.
What changed
The approach swaps semantic-class indexing for continuous, embedding-conditioned retrieval and mixture-based uncertainty modeling. It uses interaction-derived parameter estimates as mixture components rather than collapsing to a point estimate, and adds support-aware fusion with a generative prior when history is insufficient, plus optional learned embedding refinement.
Where it helps
- Robotic planning/control under uncertain surface contact properties
- Online physical parameter inference from pre-contact visual cues
- Systems requiring multimodal physical belief rather than a single estimate
- Tasks that benefit from rejecting irrelevant interaction history
- Controllers that need estimates available at real-time rates
Evidence & reproduction
Reproduction not evaluated. A linked paper or repository does not establish independent reproduction.
Limitations & open questions
The provided description does not specify the precise inverse-model training details, the form of the conditional generative prior, or how the method behaves under distribution shifts beyond the reported scenarios.
Related World Models papers
Sources
- Beyond Semantic Classes: Online Physical Parameter Estimation via Embedding-Conditioned Gaussian Mixtures ↗
arXiv · primary · Checked 2026-11-10
