🤖 NVIDIA has introduced ARDY — a framework for interactive, real-time human motion generation.
The model utilizes a hybrid representation and a two-stage transformer denoiser to ensure smoothness and controllability. ARDY supports text prompts and kinematic constraints, allowing for the creation of motion sequences lasting several minutes.
🌍 ARDY bridges the gap between accurate offline models and fast online methods, paving the way for using generative AI in humanoid robot control (e.g., Unitree G1) and interactive animation.
👤 Character animation can now be controlled "on the fly" using text or simple commands, making the process of creating digital actors more accessible.
Source 1: https://research.nvidia.com/labs/sil/projects/ardy/ Source 2: https://github.com/nv-tlabs/ardy