Community Scout digest — 29 April – 5 May 2026. Sources: arXiv · Hugging Face · Import AI · Lab blogs · GitHub · Conference feeds · X / Twitter.
1. arXiv
InterPhys: Physics-Aware Human Motion Synthesis in a Dynamic Scene (May 1) Xing, C., Mao, W., & Liu, M. arXiv:2605.01036. https://arxiv.org/abs/2605.01036 Models forces at human-object and human-scene contact points to generate physically grounded motion in interactive environments. The first physics-as-force-exchange (rather than positional constraint) synthesis approach directly relevant to partner-based somatic practice.
PhysiGen: Integrating Collision-Aware Physical Constraints for High-Fidelity Human-Human Interaction Generation (May 1) Lei, N., Li, Y.-M., Zeng, L.-A., et al. arXiv:2605.00517. https://arxiv.org/abs/2605.00517 Reduces body interpenetration in multi-person motion generation through efficient collision detection and physics constraint optimisation. Addresses a persistent artefact in duo/ensemble movement synthesis.
MG-Former: A Transformer-Based Framework for Music-Driven 3D Conducting Gesture Generation (May 1) Qiu, K., Qin, Y., Jia, T., et al. arXiv:2605.01197. https://arxiv.org/abs/2605.01197 Transformer architecture for long-range musical structure-aware gesture generation, mapping musical phrase dynamics to full-body conducting motion. Directly relevant to music-conditioned movement generation in performance contexts.
Contact Matrix: Enhancing Dance Motion Synthesis with Precise Interaction Modeling (May 6 🚩) Chen, X., Cen, Z., Pi, H., et al. arXiv:2605.04662. https://arxiv.org/abs/2605.04662 Duet dance generation framework that explicitly models contact between dancers throughout a shared phrase. Contact precision improves synchronisation beyond rhythm-alignment baselines. Directly relevant to partnered movement and ensemble choreographic generation.
2. Hugging Face Daily Papers
The Apr 29 – May 5 window showed limited motion-specific activity in HuggingFace trending; the most viewed papers in the window were concentrated in LLM reasoning and RL alignment. No motion/dance/embodied/pose papers ranked in the top 10 for this period. https://huggingface.co/papers
3. Import AI #455 (May 4)
Clark, J. (2026, May 4). Import AI 455: Automating AI research. https://importai.substack.com/p/import-ai-455-automating-ai-research
Issue theme: automated AI R&D — the trajectory from AI-assisted to fully automated research pipelines. Key data points: Claude Opus 4 achieved 2.9× speedup on optimisation benchmarks (May 2025); Opus 4.6 reached 30× by February 2026; Claude Mythos Preview reached 52× in April. OpenAI targeting an "automated AI research intern" by September 2026. No motion, embodied AI, or body movement content this issue.
4. Lab Blogs
No new posts on motion, video generation, or embodied AI from Google DeepMind or Meta AI in the Apr 29 – May 5 window identified. Kling AI (Kuaishou) released version 2.6 with a Motion Control feature enabling reference-video-driven dance synthesis — announced via X (see §7 below); no independent blog post located.
5. GitHub Trending
No motion-capture, dance, or somatic-AI repositories appeared in Python trending this week. NVIDIA/Isaac-GR00T and its ecosystem remain the most-starred embodied AI repo but showed no new release activity in this window.
6. Conference News
HuMoGen @ CVPR 2026 — 1-page abstract submission deadline: May 10 (this Sunday). https://humogen.github.io/
MOCO 2026 — Conference concluded April 25, Montpellier. ACM proceedings available. https://moco26.movementcomputing.org/
ICRA 2026 — Vienna, June 1–4. WBCD bimanual manipulation competition live. https://2026.ieee-icra.org/
7. X / Twitter
@Kling_ai — Kling 2.6 Motion Control launched this week: upload a character image + a dance/expression reference video → generate a new motion combination for that character. Practitioner-facing tool using reference-video motion transfer. https://x.com/Kling_ai/status/2007996192433520892
@oggii_0 — Demonstrated Kling 2.6 + GPT Image 2.0 to animate a movement notation sheet into a character dance. Movement sheet as conditioning input for video generation — an interesting bridge between notation systems and generative synthesis. https://x.com/oggii_0/status/2048243683057897890
@EmbodiedAIRead — Updated commentary on VLA (Vision-Language-Action) research at ICLR 2026: 164 submissions vs 9 the prior year. Landscape blog post linked covering all ICLR 2026 VLA submissions. https://x.com/EmbodiedAIRead/status/1978957826040172579
"dance technology AI" and "somatic AI" site:x.com searches returned no additional organic on-topic posts distinct from the above in the Apr 29 – May 5 window. Sponsor/Ad results excluded.