+ All Categories
Home > Documents > AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the...

AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the...

Date post: 23-Jul-2020
Category:
Upload: others
View: 1 times
Download: 0 times
Share this document with a friend
53
Pradeep Kumar Gupta | August 2017 Global Head – Solutions Architecture and Engineering-Autonomous Driving AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES PLATFORM
Transcript
Page 1: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

Pradeep Kumar Gupta | August 2017

Global Head – Solutions Architecture and Engineering-Autonomous Driving

AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES PLATFORM

Page 2: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

2

AGENDA

Autonomous Vehicles

Functional Safety

Levels of Automation

DRIVE PX Platform

Development Workflow

Page 3: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

3

NVIDIA

#1 PC Gaming #1 Pro Graphics #1 Accelerated Computing #1 AI Computing

Founded in 1993 — 11,000 Employees Invented the GPU Invented GPU-accelerated Computing

Fastest Supercomputers in Japan, U.S., Europe Pioneering AI Car Computer Nintendo Switch

Page 4: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

4

AUTONOMOUS VEHICLE

Page 5: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

6

AI REVOLUTIONIZING TRANSPORTATION

280B Miles per year 800M parking spots for 250M cars in the U.S.

Domino’s: 1M Pizzas delivered per Day

Page 6: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

7

COLLISIONS

CRITICAL REASONATTRIBUTED TO

ESTIMATED

NUMBERPERCENTAGE*

±95% CONF. LIMITS

Drivers 2,046,000 94% ±2.2%

Vehicles 44,000 2% ±0.7%

Environment 52,000 2% ±1.3%

Unknown Critical Reasons

47,000 2% ±1.4%

Total 2,189,000 100%

* Percentages are based on unrounded estimated frequencies (Data Source: NMVCCS 2005-2007)

CRITICAL REASON

ESTIMATED (Based on 94% of the NMVCCS Crashes)

NUMBERPERCENTAGE*

±95% CONF. LIMITS

Recognition Error 845,000 41% ±2.2%

Decision Error 684,000 33% ±3.7%

Performance Error 210,000 11% ±2.7%

Non-Performance Error (sleep, etc.)

145,000 7% ±1.0%

Other 162,000 8% ±1.9%

Total 2,046,000 100%

* Percentages are based on unrounded estimated frequencies (Data Source: NMVCCS 2005-2007)

Page 7: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

8

32

34

36

38

40

42

44

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

MOTOR VEHICLE FATALITY ESTIMATESNumber of deaths in thousands per year

Year

Source: National Safety Council: http://www.nsc.org/NewsDocuments/2017/12-month-estimates.pdf

Page 8: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

9

FUNCTIONAL SAFETY

“Part of overall safety relating to the equipment under control and the

control system that depends on the correct functioning of the

electrical, electronic, and programmable electronic safety-related

systems and other risk reduction measures”

–IEC 61508-4:2010; 3.1.12

“Absence of unreasonable risk due to hazards caused by malfunctioning

behavior of electrical/electronic systems”

–ISO 26262-1:2011; 1.51

Page 9: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

10

FUNCTIONAL SAFETY

Page 10: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

11

FUNCTIONAL SAFETY

ASILs identified by Standard - ASIL A, ASIL B, ASIL C, ASIL D.

ASIL D dictates the highest integrity requirements on the product and ASIL A the lowest.

ASIL’s are 3D with 3 variables: severity, probability of exposure, and controllability.

ISO 26262-3, section 7 “Hazard analysis and risk assessment” provides tables that break these three variables down into classes.

Probability of exposure has five classes: “Incredible” to “High probability” (E0-E4).

Severity has four classes: “No injuries” to “Life-threatening injuries (survival uncertain), fatal injuries” (S0-S3).

ASIL Standard

Page 11: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

12

FUNCTIONAL SAFETY

Controllability, which means controllability by the driver, not by the vehicle electronic systems, has four classes: “Controllable in general” to “Difficult to control or uncontrollable.”

http://www.electronicdesign.com/embedded/understanding-iso-26262-asils

http://www.eetimes.com/document.asp?doc_id=1331459&page_number=4

Why ASIL D now?

If your car meets the ISO 26262 specification for ASIL D, it means that the machine [car] is “making a decision for you” when it comes across a safety-critical path, explained Raucher. At ASIL B, the car would warn a driver of imminent danger, but at ASIL D, the car — faced with a hazard — brakes and stops or pulls into a safe space.”

ASIL Standard

Page 12: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

13

LEVELS OF AUTOMATION

Source: SAE International – Levels of Driving Automation

Page 13: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

15

150M AUTONOMOUS VEHICLES BY 2025

Source: ABI

2017 2018 2019 2020 2021 2022 2023 2024 2025

—130M L2

—15M L3

—5M L4

Page 14: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

17

Artificial IntelligenceAutonomous Vehicles

NVIDIA AND DEEP LEARNINGAt the Center of AV Revolution

Page 15: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

18

DEEP LEARNING AUTOMOTIVE PARTNERS

Page 16: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

19

AUTONOMOUS VEHICLES

It is quite clear that putting Autonomous Vehicles on the road is possible only with Artificial Intelligence. AV = AI.

The key challenge with Autonomous Vehicles is that we need an AI super computer in the car.

And, we brought the DRIVE Computer to address the challenge of Autonomous Vehicles

Key summary points

Page 17: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

20

DRIVE PLATFORMLevel 2 -> Level 5

1 TOPS

10 TOPS

100 TOPS

DRIVE PX 2 ParkerLevel 2/3

DRIVE PX XavierLevel 4/5

DRIVE PX 2

2 PARKER + 2 PASCAL GPU | 20 TOPS DL | 120 SPECINT | 80W

DRIVE PX (Xavier)

30 TOPS DL | 160 SPECINT | 30W

ONE ARCHITECTURE

Page 18: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

22

2 Complete AI SystemsPascal Discrete GPU

1,280 CUDA Cores4 GB GDDR5 RAM

Parker SOC Complex256 CUDA Cores4 Cortex A57 Cores2 NVIDIA Denver2 Cores8 GB LPDDR4 RAM64 GB Flash

Safety MicroprocessorAURIX Safety Microprocessor

DRIVE PX 2 COMPUTE COMPLEXES

14

Page 19: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

23

IS IT THAT SIMPLE?

Page 20: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

24

NVIDIA DRIVE – SOFTWARE

Computer Vision Libraries

OS

Perception AI

CUDA, cuDNN, TensorRT

Localization Path Planning

Page 21: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

25

DRIVE PLATFORM

DriveWorks

CUDA, TensorRT

DRIVE PX 2 — System Software

DRIVE PX 2 — System Hardware Vehicle Control

DNNs and Autonomous Driving Applications

Sensors

HD Maps

Page 22: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

26

Modular

Scalable

Optimized for GPU

Rapid prototyping & production

DESIGN PHILOSOPHY

APIs(> 450 AND GROWING)

USE CASES

DRIVEWORKS

Autonomous Driving

Data

Data Acquisition

Page 23: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

27

DRIVE PX 2

Linux PC

DEVELOP BENCHMARK/DEPLOY

Cross-Compile

Feedback & Iterate

GETTING STARTED…Workflow

Page 24: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

28

Sensors

Sensors

Trained Neural

Network Model

(DNN)

HD Maps

PC Based Algorithm /

App Development

Real / Simulated

Sensors

Real / Simulated

Vehicle

Data Acquisition &

Sensor Fusion

Data Acquisition &

Sensor Fusion

Neural Network

Training

Trained Neural

Network Model (DNN)

Trained Neural

Network Model

(DNN)

Hi Def Maps

Data Acquisition &

Sensor Fusion

Algorithms,

Applications

In-Vehicle Computing Platform

DIFFERENT WORKFLOWS

Page 25: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

29

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DEVELOPMENT FLOWSelf Driving Vehicle Development

DRIVE PX PC

Page 26: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

30

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DEVELOPMENT FLOWSelf Driving Vehicle Development

DRIVE PX PC

Page 27: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

31

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DEVELOPMENT FLOWSelf Driving Vehicle Development

DRIVE PX PC

Page 28: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

32

HOW MUCH DATA?

Page 29: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

34

TARGETED DATA COLLECTION

Learning to drive

Lane keeping, emergency braking, intersections

Perceiving your surroundings

Detect lane boundaries, drivable paths, cars, pedestrians

Summarizing your world-view to support others

Provide my understanding of this location to other vehicles

Updating the “base map”

Use Cases

Cam -60FOV

Cam- 120FOV

LIDAR

IMU/GPS

Page 30: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

35

Sensor fusion interfaces:GMSL camera, CAN, GbE, BroadR-Reach, FlexRay, LIN, GPIO

Displays interfacesHDMI, FPDlink III and GMSL

Storage interfaces10GbE, USB3

INTERFACES70 Gigabits per second of I/O

DRIVE PX 2

Tegra A (CPU+ integrated

GPU)

Pascal A (discrete GPU)

Tegra B(CPU+ integrated

GPU)

Pascal B (discrete GPU)

ASIL-D Safety

Microcontroller

CANCamera

GPIOsGigabit Ethernet

DisplayBroadR-Reach

LINUSB 3.0

FlexRayUSB 2.0

10 Gigabit EthernetGigabit Ethernet

Dispay (HDMI)JTAG

Gigabit Ethernet

Auto Grade Connectors Debug/Lab Interfaces

Page 31: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

36

FPDLinkVideo Out

GMSLVideo Out

Vehicle Harness

• 6 CAN• 4 LIN• 2 FlexRay• GbE• AURIX Debug

HDMI Parker AUSB Debug

10 GbE

HDMI Parker B

GbEUSB 3.0 Parker AUSB 3.0 Parker B

BroadR-Reach

Ethernet+ 8 GPIO

12 Cameras

• GMSL• Power over Coax

DRIVE PX 2 HARDWARE CONNECTIVITY

Page 32: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

37

DISTRIBUTED RECORDING

Multiple sensors; multiple DRIVE PX devices log synchronized data

Each DRIVE PX unit collecting data with a timestamp recording time (ms)

Data between units sync’d based on timestamps

DriveWorks Tools

Page 33: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

38

RECORDING TOOLDriveWorks Tools

Calibration & Sensor Registration

1. Route setting

2. Configuration switching

3. Storage info

4. Camera preview

5. Start Recordingcontrol button

6. Settings button7. GPS info8. Sensor Status

Page 34: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

39

CAMERA CALIBRATION & SENSOR REPLAY DriveWorks Tools

CALIBRATION & SENSOR REGISTRATION DATA REPLAY TOOL

Calibrate N

cameras

Cameras can be

pinhole or

fisheye

No restriction on

field of views

No manual

measurements

involved

Replays the sensor

data captured

Displays the data for

each sensor type

Provides a simple UI

for quick sanity checks

Page 35: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

40

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DRIVE PX PC

DEVELOPMENT FLOWSelf Driving Vehicle Development

Page 36: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

41

Page 37: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

42

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DRIVE PX PC

DEVELOPMENT FLOWSelf Driving Vehicle Development

Page 38: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

43

DRIVEWORKS MODULES

Sensing Maps

Vehicle IO*

Egomotion

Image Pipeline

Image Processing

Computer Vision

Renderer

Detection

Multi-Camera

Lidar

Radar

Map Birds-Eye View

SfM Stereo

Rectifier Color Correction

TrajectoryDNN Inference

Page 39: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

44

Page 40: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

45

AI CO-PILOT

Autonomous driving modes are OFF

Vehicle continues to be aware of its surroundings

Driver facing camera

Driver notification via:

Audio (sounds and language)

Visual (LEDs, icons, text, HUD)

Vibration (steering wheel, seat, seatbelt)

Adding Value When Humans Choose to Drive

Page 41: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

46

AI CO-PILOTConvenience

Driver Recognition

Route Memory

Driver Habit Customizations

Page 42: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

47

AI CO-PILOTCommunication

Lip Reading

Natural Language Understanding

Conversational Interaction

Page 43: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

48

AI CO-PILOTSafety

Meaningful Alerts Filter

Tracks Where Driver is Looking

Notifies on Unseen Situations

Page 44: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

49

DATA ACQUISITION

Sensor integration

Sensor calibration

Image capture

Storage

SIMULATE, DRIVE, TEST

HIL/SIL simulation

Vehicle integration

Test

DATA CURATION, LABELING

Manage data lifecycle

DNN TRAINING

Train NN with data

MAP CREATION

Create HD map

APPLICATION DEVELOPMENT

Compute libraries

NN Inferencing

System level applications

DRIVE PX PC

DEVELOPMENT FLOWSelf Driving Vehicle Development

Page 45: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

50

PX2: ON WHEELS

HARDWARE

DRIVE PX 2 nicely mounted in the trunk of a car, pre-wired for cameras, and other sensors

Sensors — Your choice or NVIDIA’s configuration

SOFTWARE

All of what comes with DRIVE PX2

Page 46: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

51

DRIVE PX + SENSORS CONFIGURED TO GO

With Ford Fusion + DRIVE PX + cameras, LIDAR, radar, navigation sensors and storage options

Photo courtesy of AutonomouStuff

Page 47: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

52

DRIVE PLATFORM

QNX — Common arch across Linux and QNX

Common APIs across CUDA, TensorRT, cuDNN, NvMedia + Support for POSIX APIs

For Production

ZF Pro AI — built on DRIVE PX 2 Auto Cruise

Bosch — building on DRIVE PX using Xavier

Xavier: 30 DL TOPS + ASIL-C for Level 4

SOFTWAREHARDWARE

Page 48: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

53

NVIDIA AI CARS

Baidu

nuTonomyVolvo WEpods

Mercedes-BenzTeslaAudi

RoboRace

Page 49: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

54

Page 50: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

55

Page 51: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

56

Page 52: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

57

SUMMARY

What are autonomous vehicles

Introduced the concept of functional safety

Discussed the levels of automation

AV & Co-Pilot

DRIVE PX

Development workflow

Page 53: AN OVERVIEW OF NVIDIA’S AUTONOMOUS VEHICLES …...“Part of overall safety relating to the equipment under control and the control system that depends on the correct functioning

QUESTIONS?


Recommended