ARMS: Anchor-Relational Motion Streaming for Seamless Solo-Social Motion Transitions

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

The ARMS (Anchor–Relational Motion Streaming) framework addresses the challenge of generating temporally continuous and socially coherent human motion from text, particularly for long-horizon streams involving solo-social transitions. This novel causal autoregressive diffusion framework unifies solo motion and human-human interaction within a single generative process. ARMS introduces a dynamics-asymmetric representation that decouples individual temporal evolution from inter-person alignment using a partner-referenced relative-translation term, ensuring seamless social coupling changes without sacrificing long-horizon stability. A causal relational diffusion model progressively refines motion segment by segment, leveraging past context and mode-aware relational gating to manage cross-agent connections for both solo and interaction modes. Experiments demonstrate ARMS significantly improves transition smoothness and social coherence, achieving lower Peak Jerk (0.071 vs. 0.265) and Area Under the Jerk (2.925 vs. 22.440) for interaction streaming compared to baselines like InterMask, and superior performance on HumanML3D (272-dim) and InterX datasets.

Key takeaway

For AI Scientists and animation developers building virtual agents or interactive environments, ARMS offers a critical advancement for generating dynamic, socially aware motion. You should consider integrating this framework to overcome the limitations of traditional fixed-length or single-mode motion generation. ARMS enables seamless transitions between solo and multi-person interactions, ensuring your characters exhibit temporally continuous and spatially consistent behaviors over long horizons, significantly enhancing realism and user experience.

Key insights

ARMS unifies solo and social motion generation for seamless, continuous human behavior streams.

Principles

• Decouple temporal evolution from inter-person alignment.
• Use partner-referenced relative-translation for coupling.
• Employ mode-aware gating for solo/interaction switching.

Method

ARMS uses a causal relational diffusion model to progressively refine motion segments using past context and mode-aware relational gating.

In practice

• Synthesize long-horizon solo-social motion streams.
• Preserve inter-person geometry and temporal coherence.
• Integrate into virtual agents or animation pipelines.

Topics

• Text-to-Motion Synthesis
• Human-Human Interaction
• Causal Diffusion Models
• Streaming Motion Generation
• Dynamics-Asymmetric Representation
• Virtual Agent Animation

Code references

• kk9six/arms

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• SiGnature: Explicit Motion Diffusion for Stylized Semantic Gesture — SiGnature reconciles precise semantic gesture control with high-fidelity speaker style preservation using explicit motion integration.
• InteractiveAvatar: Real-Time Streaming Video Generation for Consistent and Intent-Aware Avatars — InteractiveAvatar enables real-time, consistent, and intent-aware avatar video generation for infinite streaming.
• Latent Dynamics for Full Body Avatar Animation — A learned latent dynamics model evolves a residual latent for history-dependent clothing dynamics in 3D Gaussian avatars.
• Can text *finally* make robots dance exactly how we want them to? — Scaling motion generation models with high-quality data enables nuanced instruction-following capabilities.
• PhyGenHOI: Physically-Aware 4D Generation of Dynamic Human-Object Interactions — PhyGenHOI couples generative human motion with explicit physical object simulation for realistic 4D human-object interactions.
• Streaming Video Generation with Streaming Force Control — StreamForce enables real-time, physically-grounded video generation with dynamic, unified force control.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.