NVIDIA launches Ardy real-time motion model
NVIDIA's Spatial Intelligence Lab has developed Ardy, an autoregressive diffusion model for real-time, interactive 3D human motion generation. The model supports online text prompting and flexible kinematic constraints at inference time without requiring retraining, making it suitable for animation, gaming, and robotics.
Ardy bridges the gap between high-latency offline diffusion models and low-control online systems, potentially revolutionizing real-time interactive experiences and humanoid robot control.
* Bridges the gap between fast autoregressive models and high-quality diffusion techniques.
* Kinematic constraints are enforced at inference time, offering high flexibility without retraining.
* The model's real-time capabilities open new doors for interactive avatars, virtual production, gaming, and humanoid robotics.
DISCOVERED
7h ago
2026-07-19
PUBLISHED
7h ago
2026-07-19
RELEVANCE
AUTHOR
AI Search