Safety in the Loop: Open Datasets and Models for Developing Safe AVs
In this livestream, we’ll showcase how NVIDIA’s new Alpamayo open ecosystem for AV development, and in particular its AV dataset, creates new opportunities to advance AV safety. Join the livestream to learn: - How to interact with one of the largest, most geographically diverse collections of multi-sensor data (multi-camera, LiDAR, and radar) empowering AV researchers to build the next generation of end-to-end driving systems. - How to use various curation tools, from accessing metadata labels to language-based retrieval to reasoning traces, to identify scenarios of relevance to safety objectives. - Paradigms for iterative “data flywheel”-based development of safety evaluators, runtime monitors, and E2E driving models applicable to research with this dataset and beyond. Don’t miss this unique opportunity to engage with our experts directly. ---------------------------------------------------------------------------- Key Moments: 0:04 Introduction to Safety in the Loop 2:20 AlpaMayo ecosystem overview 6:02 Foundation model philosophy for AV 8:15 Cosmos world foundation models 9:49 Data flywheel concept 11:22 Physical AI AV dataset overview 16:04 Safety data flywheel explained 24:03 Dataset diversity and scale 29:02 Cosmos dataset search demo 34:52 Parallel stack safety architecture 41:18 Anonymization and privacy discussion 47:14 Auto-labeling and KPI considerations 53:00 Reasoning consistency and alignment 58:08 Counterfactual safety analysis 1:02:47 GTC workshops and ecosystem roadmap 1:06:23 Automotive sessions at GTC ---------------------------------------------------------------------------- 👥Join the Community → https://nvda.ws/3ZMfc6e ---------------------------------------------------------------------------- #AutonomousVehicles #AlpaMayo #PhysicalAI #AVSafety #ReasoningModels #VisionLanguageAction #CosmosAI #ClosedLoopSimulation #DataFlywheel #AIAlignment #RobotLearning #SelfDriving #EmbodiedAI #GTC2026 #OpenAIResearch