Today, NVIDIA earned a top spot as a winner in the Autonomous Grand Challenge at the Computer Vision and Pattern Recognition (CVPR) conference, taking place this week in Nashville, Tennessee. The recognition came during the Embodied Intelligence for Autonomous Systems on the Horizon Workshop.
This victory represents NVIDIA’s second year in a row leading the End-to-End Driving at Scale category and their third straight Autonomous Grand Challenge award at CVPR.
This year’s competition centered on “Towards Generalizable Embodied Systems,” utilizing NAVSIM v2, an advanced, data-centric simulation environment for non-reactive autonomous vehicles (AVs).
The event allowed researchers to investigate methods for managing unforeseen scenarios, extending beyond traditional real-world driving datasets, to fast-track the creation of more intelligent and secure AV technologies.
Creating Secure and Flexible Driving Paths
Challenge participants focused on producing driving paths from multi-sensor inputs within a semi-reactive simulation setup, where the primary vehicle’s route is predetermined, while surrounding traffic evolves in real time.
Entries were assessed via the Extended Predictive Driver Model Score, evaluating factors like safety, passenger comfort, adherence to rules, and adaptability across both real and simulated environments—advancing the field of reliable, versatile autonomous driving.
The NVIDIA AV Applied Research Team introduced the Generalized Trajectory Scoring (GTRS) approach, a standout method that produces multiple trajectory options and iteratively selects the optimal one.
GTRS employs a blend of broad trajectory collections to address diverse conditions and precise paths for high-risk moments, generated via a diffusion policy tailored to the surroundings. It then applies a transformer decoder, refined from metrics reliant on perception, emphasizing safety, ease, and regulatory compliance. This decoder systematically narrows down top trajectory choices by detecting nuanced yet vital distinctions among comparable options.
The framework demonstrates strong versatility across numerous scenarios, delivering cutting-edge performance on demanding tests and supporting dependable, responsive path choices in varied, complex driving situations.
NVIDIA’s Automotive Innovations at CVPR
Over 60 papers from NVIDIA were selected for CVPR 2025, covering areas such as automotive, healthcare, robotics, and beyond.
In the automotive domain, NVIDIA’s team is pioneering physical AI through advancements in perception, planning, and data synthesis. Notably, three papers from NVIDIA received Best Paper Award nominations: FoundationStereo, Zero-Shot Monocular Scene Flow, and Difix3D+.
The following NVIDIA contributions highlight progress in stereo depth estimation, monocular motion analysis, 3D reconstruction, closed-loop planning, vision-language integration, and generative simulations—all essential for developing more secure and adaptable AVs:
Discover automotive-focused workshops and sessions at CVPR, including:
- Workshop on Data-Driven Autonomous Driving Simulation, with Marco Pavone, senior director of AV research at NVIDIA, and Sanja Fidler, vice president of AI research at NVIDIA
- Workshop on Autonomous Driving, featuring Laura Leal-Taixe, senior research manager at NVIDIA
- Workshop on Open-World 3D Scene Understanding with Foundation Models, featuring Leal-Taixe
- Safe Artificial Intelligence for All Domains, featuring Jose Alvarez, director of AV applied research at NVIDIA
- Workshop on Foundation Models for V2X-Based Cooperative Autonomous Driving, featuring Pavone and Leal-Taixe
- Workshop on Multi-Agent Embodied Intelligent Systems Meet Generative AI Era, featuring Pavone
- LatinX in CV Workshop, featuring Leal-Taixe
- Workshop on Exploring the Next Generation of Data, featuring Alvarez
- Full-Stack, GPU-Based Acceleration of Deep Learning and Foundation Models, led by NVIDIA
- Continuous Data Cycle via Foundation Models, led by NVIDIA
- Distillation of Foundation Models for Autonomous Driving, led by NVIDIA
Check out the NVIDIA research papers slated for presentation at CVPR, and view the NVIDIA GTC Paris keynote delivered by NVIDIA founder and CEO Jensen Huang.
Discover more about NVIDIA Research, a worldwide group of hundreds of experts dedicated to fields like AI, computer graphics, computer vision, self-driving vehicles, and robotics.
The lead image depicts an AV adjusting its path to maneuver through a cityscape with shifting traffic, powered by the GTRS model.
