Part-level control, biologically-inspired two-tier architectures, and real-time humanoid motion move into the spotlight

From the arXiv

FrankenMotion: Part-Level Human Motion Generation and Composition (arXiv:2601.10909) Li et al. enable fine-grained, asynchronous per-body-part control in text-to-motion, backed by FrankenStein — a new dataset with atomic, temporally-aware part-level annotations produced by an LLM agent (93% expert-validated). Hierarchical conditioning across part/action/sequence levels; state-of-the-art on part-based generation. Raises the somatic question of composition (parts) vs. integration (whole). → https://arxiv.org/abs/2601.10909 · https://coral79.github.io/frankenmotion/

OMG: Omni-Modal Motion Generation for Generalist Humanoid Control (arXiv:2606.10340) Tsinghua MARS Lab decomposes humanoid control into a generative "brain" (OMG-DiT, reasoning over language + audio + reference motion) and a reactive tracking "cerebellum," explicitly mirroring biological motor hierarchy. Built on 1,000+ hours of omni-modal data; runs in real time on a Unitree G1. Code + project page public. → https://arxiv.org/abs/2606.10340 · https://tsinghua-mars-lab.github.io/OMG/

BRIC: Bridging Kinematic Plans and Physical Control at Test Time (arXiv:2511.20431, AAAI 2026) Lim et al. reconcile a diffusion kinematic planner with an RL physics controller via test-time adaptation — no retraining — keeping intended movement and physical execution aligned over long horizons. Addresses execution drift, a key barrier to extended-duration generation. → https://arxiv.org/abs/2511.20431

From the Industry

Tencent / MoreFun Studios — real-time AI Kung Fu (GDC 2026) — At the Tencent Games AI Summit, MoreFun Studios' head of AI presented a real-time AI technique generating fluid Kung Fu motion for the game The Hidden Ones. Notable as a production-deployment signal: real-time AI motion generation is moving from research into shipping interactive entertainment, where responsiveness and movement quality both matter to players. The martial-arts domain is one where movement quality (not just correctness) is central — a commercial proving ground for quality-aware generation.

From Import AI

Import AI #464 (Jack Clark, early July 2026) — Continuing threads on the measurable acceleration of AI capability: the "Remote Labor Index" success rate rose from 2.5% at its October 2025 launch to 16.1% by July 2026, and Anthropic reported an ~8× increase in code merged into its codebase in 2026 versus 2021–2024. The issue examines what these compounding numbers mean for the pace of AI integration into real work. (Less movement-specific this week; relevant as macro-context on how fast the surrounding field is moving.) → https://importai.substack.com/

From X / Social

@MoreFunStudios (Tencent) — promoted the GDC 2026 real-time Kung Fu motion generation talk, with clips showing fluid, responsive martial-arts motion generated live. Drew interest from the game-dev and motion-AI communities on the responsiveness/quality trade-off in real-time generation.

Tsinghua MARS Lab — released OMG demo videos showing the Unitree G1 executing language-, audio-, and reference-conditioned motion in real time. The brain/cerebellum framing (generative reasoning atop reactive tracking) resonated with researchers interested in biologically-grounded control hierarchies — the same intermediate-layer theme the field has converged on through June–July.

Conference & Community Notes

Emerging emphasis — hierarchy and real-time: This week's work (OMG's brain/cerebellum, BRIC's plan/control bridging) consolidates the intent-layer/intermediate-layer theme into explicit two-tier architectures, now running in real time on physical robots. The field is not only accepting that movement needs an intermediate organising layer — it is building that layer as a distinct architectural tier.

Venues:

  • MOCO'26 (Movement and Computing) — concluded 23–25 April 2026, Cité des Arts, Montpellier; proceedings published in both tracks: ACM DL (doi:10.1145/3802842) and the Open-track book of abstracts (HAL hal-05665489 / Zenodo doi:10.5281/zenodo.20794033, CC BY 4.0). Priority read — the most thesis-aligned venue.
  • ECogS 2026 (OIST, Nov 9–13) — "Embodied cognition and AI."
  • NeurIPS 2026 — notifications September.

Continue through the archive

For connected context, read This Week in Motion AI: Parts, Wholes, and the Brain–Cerebellum Split and Can You Build a Movement Out of Parts? What AI's Newest Trick Reveals About the Body.

References

Li, C., et al. (2026). FrankenMotion: Part-level human motion generation and composition. arXiv:2601.10909. https://arxiv.org/abs/2601.10909

Lim, D., Kim, M., Lim, J., & Kim, S. (2026). BRIC: Bridging kinematic plans and physical control at test time. arXiv:2511.20431. https://arxiv.org/abs/2511.20431

Tsinghua MARS Lab. (2026). OMG: Omni-modal motion generation for generalist humanoid control. arXiv:2606.10340. https://arxiv.org/abs/2606.10340