The clearest signal in late July 2026 is architectural rather than empirical: three separate research groups published systems that deliberately remove a fixed assumption — about skeletal structure, about sensor configuration, or about body plan. EquiFusion (ECCV 2026) drops the fixed skeleton, WHIP drops the fixed sensor rig, and Two2Four (DisneyResearch|Studios) drops the human body plan entirely. This digest records what was published, by whom, and where to verify it.

Scope note: this digest covers verifiable publications and official releases from 8–28 July 2026. Items with no traceable primary source have been omitted rather than summarised vaguely.

Papers

EquiFusion: Kinematics-Agnostic Human Motion Prediction via Equivariant Latent Diffusion — Curreli, Hofherr, Muhle, Saroha, Marin & Cremers (Technical University of Munich). arXiv:2607.10984, 13 July 2026; accepted to ECCV 2026. A latent diffusion model with permutation-equivariant architecture that treats skeletal connectivity as an explicit input, enabling cross-dataset generalisation to unseen kinematics, zero-shot prediction from occluded input, and targeted limb generation. Reported as up to 75% more compact than kinematics-specific methods. Code repository is public. → https://arxiv.org/abs/2607.10984

Towards Real-World Wearable Motion Reconstruction — arXiv:2607.09780, 8 July 2026. Argues motion capture should prioritise unobtrusive consumer devices (smartphones, smartwatches, smart glasses, smart insoles). Contributes a large-scale multimodal dataset synchronising consumer sensors with ground-truth 3D motion across 50 activities, and WHIP, a generative model reconstructing motion from arbitrary sensor subsets while handling missing modalities. → https://arxiv.org/abs/2607.09780

Two2Four: Generative Quadruped Puppeteering from Human Motion — Zargarbashi, Qiu, Agrawal, Coros, Sumner, Guay & Buhmann (DisneyResearch|Studios Switzerland, ETH Zürich). arXiv:2607.26108, 28 July 2026. Two-stage diffusion trained purely on quadruped data, with structured conditioning and inpainting, maps human motion to quadruped motion across walking, running, jumping, sitting and lying, with head and individual-limb control. → https://arxiv.org/abs/2607.26108 · Project page: https://studios.disneyresearch.com/2026/07/30/two2four-generative-quadruped-puppeteering-from-human-motion/

Social Structure Matters in 3D Human-Human Interaction Generation — Wang, Z., Wang, B., Bian, Y., Wang, P., Wang, Z., Dong, D., Li, H., Mo, H., & Sun, Z. arXiv:2606.24255. Formulates text-driven two-person interaction generation as social structure modelling and grounding, rather than treating two bodies as independently generated and then combined. Relevant to partnered movement work, where the relationship between movers is the primary material rather than a post-hoc constraint — the premise underlying the Lab's proposal for an anticipatory movement partner. → https://arxiv.org/abs/2606.24255

Why these cluster

Each of these systems removes a constraint that had been treated as fixed. EquiFusion removes the fixed skeleton; WHIP removes the fixed sensor set; Two2Four removes the shared body plan; the social-structure work removes the assumption that an interaction is two solo motions placed side by side.

The common effect is that the model no longer requires the world to arrive in a predetermined format. This is worth noting for movement practice specifically, because practice conditions are variable by nature — different bodies, partial visibility, whatever sensors are at hand, and partnering where the relationship is the material. Architectures that assume a fixed configuration have historically been the reason research systems fail outside the lab.

What none of these changes is the sensing modality. All four operate on positional or inertial data. The dimension this platform tracks — movement as felt and muscularly produced, examined in the analysis of muscle redundancy as the physical substrate of movement quality — remains outside their scope.

Venues

  • MOCO'26 (Movement and Computing) — concluded 23–25 April 2026, Cité des Arts, Montpellier. Proceedings published in two tracks: ACM Digital Library (doi:10.1145/3802842) and an Open-track book of abstracts (HAL hal-05665489 / Zenodo doi:10.5281/zenodo.20794033, CC BY 4.0).
  • ECCV 2026 — EquiFusion accepted; conference proceedings forthcoming.
  • ECogS 2026 (International Conference on Embodied Cognitive Science, OIST) — 9–13 November 2026, theme "Embodied cognition and AI."

References

Curreli, C., Hofherr, F., Muhle, D., Saroha, A., Marin, R., & Cremers, D. (2026). EquiFusion: Kinematics-agnostic human motion prediction via equivariant latent diffusion. arXiv:2607.10984. https://arxiv.org/abs/2607.10984

Towards real-world wearable motion reconstruction. (2026). arXiv:2607.09780. https://arxiv.org/abs/2607.09780

Wang, Z., Wang, B., Bian, Y., Wang, P., Wang, Z., Dong, D., Li, H., Mo, H., & Sun, Z. (2026). Social structure matters in 3D human-human interaction generation. arXiv:2606.24255. https://arxiv.org/abs/2606.24255

Zargarbashi, F., Qiu, Z., Agrawal, D., Coros, S., Sumner, R. W., Guay, M., & Buhmann, J. (2026). Two2Four: Generative quadruped puppeteering from human motion. arXiv:2607.26108. https://arxiv.org/abs/2607.26108