Welcome to the homepage of NVIDIA’s Data-Driven AI for Robotics (DAIR) group, led by Umar Iqbal. We are part of the Learning and Perception Research (LPR) organization within NVIDIA Research.
Our group investigates how robots can learn directly from human data, such as videos, motion capture, and large-scale demonstrations, to acquire skills that generalize across tasks, embodiments, and environments. We work at the intersection of computer vision, machine learning, and robotics, developing models that understand, reconstruct, and imitate human behaviors.
Our research contributes to NVIDIA’s broader vision of foundation models for robotics, combining advances in human motion modeling, human–object and human–scene interaction modeling, physics-based simulation, and embodied intelligence to enable scalable robot learning. Ultimately, we aim to bridge the gap between human understanding and robotic intelligence, advancing the goal of robots that learn by watching humans.
Computer Vision, Machine Learning, Generative AI
3D Vision, Embodied AI, Reinforcement Learning
Computer Vision, 3D Vision
3D Vision, Robotics, Human Motion
Emboddied Intelligence, 3D Scene Understanding, Human Modeling
Computer Graphics, Generative AI, 3D Vision