Post Content
Generalist robots are here, but training them effectively poses a challenge due to the scarcity of both real and synthetic data, as well as the difficulty of scaling human demonstrations.
#NVIDIAIsaac GR00T-Dreams, a cutting-edge blueprint built on #NVIDIACosmos, generates large-scale synthetic trajectory data from minimal human demonstrations, enabling robots to quickly learn a wide array of new actions.
Cosmos Predict, fine-tuned with single-task teleoperation data, generates 2D synthetic “”dream”” videos of potential future states. Cosmos Reason then evaluates and selects the optimal dreams, from which the blueprint creates 3D action trajectories for robot training.
This innovative approach brings developers one step closer to overcoming the challenge of robot data.
Learn more about GR00T-Dreams: https://nvda.ws/4kuqGWu
Learn more about Isaac GR00T: https://nvda.ws/4mnOSeO
Learn more about Cosmos Reason: https://nvda.ws/4jY7b8B
#HumanoidRobots #PhysicalAI #NVIDIAAI #AI #COMPUTEX2025 Read More NVIDIA
#Techno #nvidia
