END TO END NEEDS FOR AUTONOMOUS VEHICLES NORM MARKS SEPT. 6, 2018
THE MOST EXCITING TIME IN TECH HISTORY GAMING $100B Industry ARTIFICIAL INTELLIGENCE $3T IT Industry AUTONOMOUS VEHICLES $10T Transportation Industry NVIDIA GPU 2
END-TO-END SYSTEM FOR AV Cars Pedestrians Path Lanes Signs Lights COLLECT DATA TRAIN MODELS SIMULATE RE-SIMULATE MAPPING Sources: NVIDIA & RAND Corporation 3
NVIDIA PERCEPTION INFRASTRUCTURE LARGE-SCALE DEEP LEARNING MODEL DEVELOPMENT Data Factory Train on NVIDIA DGX Library of Labeled Data Workflow, Tools, Supercomputing Infrastructure Data Ingest, Labeling, Training, Validation, Adaptation Automation, Best Model Discovery, Traceability, Reproducibility Purpose-built for Safety Standards of Automotive Data is the new source code DRIVE Pegasus Validate/ Verify Test Data 4
DATA COLLECTION AND LABELING FOR AI 100 s of petabytes of data from test vehicles 20% to 50% of data may not be useful 10+ DNNs for self-driving vehicles 10 s of billions of total images from test vehicles 1,500 workers Label up to 1M images per month Raw Data Total Images Useful Data Labeled Images # DNNs Source: Data from test fleets of 50-100 cars 5
AI FOR SELF-DRIVING WORKFLOW Get Data Train & Test Adjust Deploy Test & Validate Labeled Data Trained Model Fine Tune Model Export Model Inference at Edge DNN Development Exploration Development Model Selection Simulate Re-Simulate 6
AI FOR SELF-DRIVING 7
8
AI OUTSIDE AND INSIDE THE VEHICLE Exterior Driver Recognition Automatic Personalization Device usage detection Cyclist Alert Distracted Driver Alert Driver/Passenger Recognition Customer Application DRIVE AV Object, Freespace, Path / Lane, Path Planning, Wait, Map, Sign, Lights, Road Markings, Surround DRIVE IX Gaze, Eye Openness, Head Pose, Gestures, Emotions Facial Recognition, Voice Recognition & Lip Reading DRIVE OS 9
MANY THINGS TO LEARN 10
Autonomous vehicles need to be driven more than 11 billion miles to be 20% better than humans. With a fleet of 100 vehicles, 24 hours a day, 365 days a year, at 25 miles per hour, this would take 518 years. Rand Corporation, Driving to Safety 11
SIMULATION THE PATH TO BILLIONS OF MILES World drives trillions of miles each year. U.S. has 770 accidents per billion miles. A fleet of 20 test cars cover 1 million miles per year. 12
NVIDIA DRIVE SIM AND CONSTELLATION AV VALIDATION SYSTEM Virtual Reality AV Simulator Same Architecture as DRIVE Computer Simulate Rare and Difficult Conditions, Recreate Scenarios, Run Regression Tests, Drive Billions of Virtual Miles 1,000 s Constellations Drive Billions of Miles per Year 13
NVIDIA DRIVE SIM AND CONSTELLATION AV VALIDATION SYSTEM Virtual Reality AV Simulator Same Architecture as DRIVE Computer Simulate Rare and Difficult Conditions, Recreate Scenarios, Run Regression Tests, Drive Billions of Virtual Miles 1,000 s Constellations Drive Billions of Miles per Year 14
NVIDIA DRIVE SIM AND CONSTELLATION AV VALIDATION SYSTEM Virtual Reality AV Simulator Same Architecture as DRIVE Computer Simulate Rare and Difficult Conditions, Recreate Scenarios, Run Regression Tests, Drive Billions of Virtual Miles 1,000 s Constellations Drive Billions of Miles per Year 15
NVIDIA DRIVE SIM AND CONSTELLATION AV VALIDATION SYSTEM Virtual Reality AV Simulator Same Architecture as DRIVE Computer Simulate Rare and Difficult Conditions, Recreate Scenarios, Run Regression Tests, Drive Billions of Virtual Miles 1,000 s Constellations Drive Billions of Miles per Year 16
MULTI-SENSOR SIMULATION 17
NVIDIA DRIVE END-TO-END PLATFORM COLLECT & PROCESS DATA TRAIN MODELS Cars Pedestrians Lanes Path Signs Lights SIMULATE DRIVE Cars Pedestrians Lanes Path Signs Lights 18
CARS TRUCKS 370 PARTNERS DEVELOPING ON NVIDIA DRIVE MOBILITY SERVICES SUPPLIERS MAPPING LIDAR CAMERA / RADAR STARTUPS 19
KEY TAKEAWAYS 1. Understand end-to-end requirements of autonomous vehicle development 2. AI demands data center design built on dense GPU compute-at-scale 3. Consider the complete workflow of AI from experimentation to training to inference 4. Carefully weigh cost of productivity vs hardware cost alone = true TCO of DL 5. NVIDIA best practices leads to TSTADI reference platform (Training, Simulation, Testing for Autonomous Driving Infrastructure) 20
THANK YOU NORM MARKS NMARKS@NVIDIA.COM