NVIDIA Research Unlocks Advanced Grasping, Smarter Autonomous Driving and Agent Training at Scale
NVIDIA , Wednesday, June 3rd, 2026
NVIDIA Research presents three foundation models for robotic grasping, autonomous vehicles, and embodied agents at CVPR.
At the Computer Vision and Pattern Recognition conference, NVIDIA Research unveiled three papers demonstrating how training at scale enables AI systems to generalize across diverse applications. GraspGen-X is the first foundation model for zero-shot robotic grasping trained on billions of simulated grasps to work with any gripper.
LCDrive introduces a compact latent representation model that allows autonomous vehicles to reason faster on embedded hardware by replacing text-based chain-of-thought with compressed spatial representations. NitroGen extends the GR00T foundation model architecture to train embodied agents in virtual environments across tens of thousands of hours of interaction, demonstrating significant performance improvements in low-data conditions.