Community Scout digest — 22–28 April 2026. Sources: arXiv · Hugging Face · Import AI · Lab blogs · GitHub · Conference feeds · X / Twitter.
1. arXiv
IAM: Identity-Aware Human Motion and Shape Joint Generation (April 28) Jia et al., 13 authors. arXiv:2604.25164. https://arxiv.org/abs/2604.25164 Models the relationship between body morphology and motion dynamics for text-driven, identity-consistent motion synthesis. Directly relevant to personalised motion priors and culturally diverse body representation in somatic AI pipelines.
HuM-Eval: A Coarse-to-Fine Framework for Human-Centric Video Evaluation (April 28) Zhang, B., Guan, K., & Song, R. arXiv:2604.25361. https://arxiv.org/abs/2604.25361 Combines Vision-Language Model global assessment, 2D pose anatomical correctness scoring, and 3D motion stability analysis for evaluating generated human video — achieving 58.2% average human correlation. Fills a gap in evaluation methodology for motion video generation that relies on human-aligned, multi-level metrics.
AnimateAnyMesh++: A Flexible 4D Foundation Model for High-Fidelity Text-Driven Mesh Animation (April 29) Wu, Z., Yu, C., Wang, F., & Bai, X. arXiv:2604.26917. https://arxiv.org/abs/2604.26917 Feed-forward text-driven animation of arbitrary 3D meshes, trained on the DyMesh-XL dataset (300K unique identities). Supports variable-length sequences for extended animations. Relevant to character animation and performance documentation in non-standard body forms.
Lifting Embodied World Models for Planning and Control (April 28) Wang, A. N., Darrell, T., Izmailov, P., Bai, Y., & Bar, A. arXiv:2604.26182. https://arxiv.org/abs/2604.26182 Composes frozen video world models with lightweight policies that map high-level actions to low-level joint sequences for human-like embodiments, achieving 3.8× lower mean joint error versus baselines. A step toward body-aware world models that could inform somatic–AI interaction design.
2. Hugging Face Daily Papers
PhyCo: Learning Controllable Physical Priors for Generative Motion (trending this week) https://huggingface.co/papers/2604.28169 — Physics-supervised fine-tuning on 100K+ photorealistic simulation videos with varying friction and force parameters, enabling "continuous, interpretable, and physically grounded control" of video generation. Relevant to physically plausible movement synthesis.
MoCapAnything V2: End-to-End Motion Capture for Arbitrary Skeletons (79 votes) https://huggingface.co/papers/2604.28130 — End-to-end learnable pipeline for joint rotation recovery from monocular video on unseen skeletal structures; 20× faster than mesh-based approaches. High community interest indicates growing demand for flexible, real-time capture.
3. Import AI Newsletter
No new Import AI issue was published in the April 22–28 window. The most recent issue, #454 (April 20), covered automated alignment research and safety evaluation — no motion or embodied AI content. The next issue is expected April 27 or May 4. https://importai.substack.com
4. Lab Blogs
No new posts in the April 22–28 window from Google DeepMind or Meta AI specifically on motion, video generation, or embodied AI. DeepMind's ANCESTRA / Veo motion matching story (director Eliza McNitt, collaboration with Primordial Soup) continues to circulate — original post dated June 2025 🚩. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/ancestra-behind-the-scenes/
5. GitHub Trending
No motion-capture, dance, or somatic-AI repositories appeared in Python trending this week. NVIDIA/Isaac-GR00T continues in the broader embodied AI conversation but was already covered in previous digests.
6. Conference News
HuMoGen @ CVPR 2026 — 1-page abstract submissions open until May 10. Third edition of the Human Motion Generation workshop; speakers from Meta Reality Labs, Peking University, ENPC, and MPI Informatics. Denver, June 3–7. https://humogen.github.io/
ECogS 2026 — International Conference on Embodied Cognitive Science, OIST, Okinawa. Theme: Embodied Cognition and AI. Dates: November 9–13, 2026. Call for abstracts open. https://x.com/DrTomFroese/status/2026992275742761202
ICRA 2026 — Vienna, June 1–4. WBCD (What Bimanuals Can Do) real-world bimanual manipulation competition. Full programme live. https://x.com/AgilexRobotics/status/2005914133527777697
7. X / Twitter
@EmbodiedAIRead — VLA (Vision-Language-Action) research grew from 9 to 164 ICLR submissions in one year; post links to a landscape blog covering all ICLR 2026 VLA submissions. Relevant signal for how embodied AI research is scaling. https://x.com/EmbodiedAIRead/status/1978957826040172579
@DrTomFroese — Keynote speakers confirmed for ECogS 2026 (Marieke Vugt, Hiro Iizuka, Jing He). Call for abstracts still open. Theme of embodied cognition + AI directly intersects with somatic practice scholarship. https://x.com/DrTomFroese/status/2026992275742761202
@MagicMotionAI — #VibeJam 2026 contest: create game animations using motion diffusion tools, share behind-the-scenes, tag @MagicMotionAI. Free 1-year Creator Plan prize. Signals growing practitioner-facing motion diffusion tooling. https://x.com/MagicMotionAI/status/2043386916138053971
"dance technology AI" site:x.com search returned no organic, on-topic posts within the April 22–28 window distinct from those above. Sponsor and off-topic results excluded.