SLIDE 1
1
2
LIVE IN THE CAR: INSIDE LOOK INTO AUTONOMOUS VEHICLE DRIVE MISSIONS - - PowerPoint PPT Presentation
LIVE IN THE CAR: INSIDE LOOK INTO AUTONOMOUS VEHICLE DRIVE MISSIONS - - PowerPoint PPT Presentation
LIVE IN THE CAR: INSIDE LOOK INTO AUTONOMOUS VEHICLE DRIVE MISSIONS Neda Cvijetic 1 WHAT ARE DRIVE MISSIONS? DRIVE MISSIONS = PIECES OF AV SW FUNCTIONALITY THAT TEACH THE CAR TO DRIVE ITSELF 2 DRIVE MISSIONS: PERCEPTION 3 WHAT A HUMAN SEES
SLIDE 2
SLIDE 3
3
DRIVE MISSIONS: PERCEPTION
SLIDE 4
4
WHAT A HUMAN SEES
SLIDE 5
5
WHAT THE AI CAR COMPUTER SEES
One slide explanation of what DNNs are and how they work, highlighting relevance of matrix multiplication
SLIDE 6
6
HOW A DEEP NEURAL NETWORK SEES
Image “Audi A7”
Image source: “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks” ICML 2009 & Comm. ACM 2011. Honglak Lee, Roger Grosse, Rajesh Ranganath, and Andrew Ng.
SLIDE 7
7
OBSTACLE PERCEPTION
SLIDE 8
8
PATH PERCEPTION
SLIDE 9
9
SLIDE 10
10
EXAMPLE: INTERSECTION DETECTION
AI-based Wait Conditions Detection
SLIDE 11
11
EXAMPLE: INTERSECTION DETECTION
AI-based Wait Conditions Detection
SLIDE 12
12
WAITNET
SLIDE 13
13
WAITNET
SLIDE 14
14
DRIVE MISSIONS: LOCALIZATION + MAPPING
SLIDE 15
15
SLIDE 16
16
SLIDE 17
17
DRIVE MISSIONS: PLANNING
SLIDE 18
18
SLIDE 19
19
NVIDIA SAFETY FORCE FIELD
https://www.nvidia.com/en-us/self-driving-cars/safety-force-field/
SLIDE 20
20
SLIDE 21
21
SLIDE 22
22
SLIDE 23
23
SLIDE 24
24
DRIVE MISSIONS: SIM + CONSTELLATION
SLIDE 25
25
DRIVE MISSIONS IN DRIVE SIM
Nightly battery of tests ( ex: Euro NCAP )
Euro NCAP CCRs: Slower target vehicle Heyford Test Track, UK
Reproducible results
SAME weather SAME time of day SAME test drivers SAME test pedestrians
SLIDE 26
26
SIM = RAPID PROTOTYPING
Dangerous or Expensive scenarios Permutations of scenarios Internationalization Different sensor configurations
Validate functionality over 1,000s of conditions
SLIDE 27
27
SIM = RAPID PROTOTYPING
Dangerous or Expensive scenarios Permutations of scenarios Internationalization Different sensor configurations
Validate functionality over 1,000s of conditions
SLIDE 28
28
“ 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
SLIDE 29
29
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 10,000 Constellations Drive 3B Miles per Year
SLIDE 30
30
AUTONOMOUS VEHICLE SENSORS
Diversity and Redundancy in Data
SLIDE 31
31
DRIVE AGX XAVIER
World’s First Autonomous Machine Processor
Most Complex SOC Ever Made | 9 Billion Transistors, 350mm2, 12nFFN | ~8,000 Engineering Years Diversity of Engines Accelerate Entire AV Pipeline | Designed for ASIL-D AV
SLIDE 32
32
NVIDIA DRIVE AGX PEGASUS
320 TOPS for AI Inferencing Memory Bandwidth >1 TB/s 400W Data Center in the Trunk
SLIDE 33
33
DRIVE MISSIONS: CULTURE OF SAFETY
SLIDE 34
34
A CULTURE OF SAFETY
One slide explanation of what DNNs are and how they work, highlighting relevance of matrix multiplication
SLIDE 35
35
A CULTURE OF SAFETY
Commitment to Recognized Industry Standards
- Commitment to all
recognized safety standards
- Commitment to
cybersecurity
- Commitment to AI education
SLIDE 36
36