NVIDIA's MotionBricks generates character motion at 15,000 FPS with 2ms latency on an RTX 5090
kimmonismus · x · 2026-10-28
NVIDIA unveiled MotionBricks (SIGGRAPH 2026), a large-scale real-time motion generation framework:
- A single neural backbone trained on 350,000+ motion clips, hitting 15,000 FPS with 2ms latency on an RTX 5090
- 'Smart primitives' provide a unified, intuitive interface for authoring navigation and object interaction
- Zero-shot transfer to new downstream tasks — no fine-tuning or task-specific tagging — enabling plug-and-play assembly like stacking bricks
- A 2:40 uncut UE5 demo shows every motion generated by neural networks: no foot-locking, blending, collision detection, or hand-authored transitions
The team says it's the first neural model to reach this level of motion quality, controllability and completeness, targeting game characters and humanoid robots.
More from Embodied
- Insta360 X6 Spatial Capture Turns a 360° Walk Into an Explorable Gaussian Splat Scene — Aiden_Tech_Ai · 2026-10-28
- Purdue's GPS-Free Drone Tracks People Through Dense Forest With $1.5M Army Funding — Graham_dePenros · 2026-10-28
- QNX expands from millions of vehicles to robotics and physical AI — pdamodaran · 2026-10-28
- ETH Zurich: foothold prediction cuts humanoid locomotion training to ~1/3 iterations — breadli428 · 2026-10-28
- VLA Insiders: Fast Policies Are 90% About Skipping Teleop Data — Kinematics Helps Too — andrew_n_carr · 2026-10-28
- UniPred, accepted at IEEE T-RO 2026, invents task concepts from demonstrations for long-horizon robot planning — tomssilver · 2026-10-28