Nvidia

NVIDIA researchers have revealed EgoScale, a framework that leverages a massive 20,854-hour egocentric human dataset to train robots in complex, fine-grained manipulation with minimal robot-in-the-loop data.

NVIDIA has released SONIC, a generalist humanoid controller trained on 100 million frames of motion data, aiming to replace manual reward engineering with a scalable "System 1" foundation for whole-body movement.

NVIDIA GEAR Lab has released DreamDojo, an open-source world model pretrained on a massive 44,000-hour dataset of human egocentric videos. By using "latent actions" to bridge the gap between human and robot movement, the model achieves zero-shot generalization and real-time controllability for teleoperation and planning.

NVIDIA GEAR Lab has unveiled DreamZero, a 14-billion parameter World Action Model (WAM) that uses video diffusion to grant robots physical "imagination," enabling zero-shot task completion and rapid adaptation across different robotic embodiments.

New research from NVIDIA and UC Berkeley demonstrates a vision-only policy that outperforms human teleoperation, cementing the Unitree G1 as the standard for sim-to-real research.