DRIVE PX / CUDA MAP MODULE

車載用ADAS/自動運転プラットフォームDRIVE PX
及びコックピト・プラットフォームDRIVE CXのご紹介
シニア・ソリューションアーキテクト 馬路 徹, 2015年9月18日
Autonomous Driving System Architecture and
DRIVE PX/CX Implementations
Agenda
DRIVE PX for Map Module, AI Module and
Computer Vision
DRIVE CX for HMI Module
Summary
2
Autonomous Driving System Architecture
Typical Architecture
白線距離
走行環境センシングおよび障害物認識
-
前方の障害物センシング(ミリ波レーダ、レーザレーダ、カメラ)
レーンマーカセンシング
エンジン・ブレーキ
速度制御モジュール
障害物
位置等
修正 - Adaptive Cruise Control
指示 - Pre-Crush System
地図モジュール
測位
GPS
車間
距離
- 固定道路地図
- ローカルダイナミックマップ
- 目標走行軌跡生成
人工知能モジュール
- 環境理解
- 判断
- 目標走行軌跡修正
交通情報等
ビッグデータ、道路・交通情報等(車外データ)
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
ハンドル
操舵制御モジュール
(車線維持制御)
道路線形
道路地図
修正
指示
HMI モジュール
-手動、自動切換え操作システム
- 稼動状況表示
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
3
Autonomous Driving System Architecture
Typical Architecture
Driving Environment Sensing and Obstacle Recognition
-
Front Obstacles Sensing (Mili-wave Radar, Laser Radar, Camera)
Lane Marker Sensing
Obstacle
Location
Position
Sensing
GPS
Lane Distance
Car
Distance
Adjusting
Acceleration
MAP MODULE
-
Road Map
Local Dynamic Map
Target Path Generation
AI MODULE
-
SPEED CONTROL
MODULE
-
Direction
Traffic Information
STEERING CONTROL
MODULE
-
Lane Keep Control
HMI MODULE
Big Data, Road, Traffic Information etc
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
Adaptive Cruise Control
Pre-Crush System
Steering
Environment Recognition
Decision Making
Target Path Tuning
Adjusting
Road Structure
Road Map
Engine, Break
- Auto/Manual Mode SW
Operation
- System Operation Status
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
4
Autonomous Driving System Architecture
MAP MODULE implementation by DRIVE PX/CUDA
Driving Environment Sensing and Obstacle Recognition
-
Front Obstacles Sensing (Mili-wave Radar, Laser Radar, Camera)
Lane Marker Sensing
Obstacle
Location
DRIVE PX / CUDA
-
Road Map
Local Dynamic Map
Target Path Generation
AI MODULE
-
Engine, Break
SPEED CONTROL
MODULE
-
Direction
Traffic Information
STEERING CONTROL
MODULE
-
Lane Keep Control
HMI MODULE
Big Data, Road, Traffic Information etc
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
Adaptive Cruise Control
Pre-Crush System
Steering
Environment Recognition
Decision Making
Target Path Tuning
Adjusting
Road Structure
Road Map
Car
Distance
Adjusting
Acceleration
MAP MODULE
Position
Sensing
GPS
Lane Distance
- Auto/Manual Mode SW
Operation
- System Operation Status
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
5
Autonomous Driving System Architecture
+ AI MODULE implementation by DRIVE PX/DL
Driving Environment Sensing and Obstacle Recognition
-
Front Obstacles Sensing (Mili-wave Radar, Laser Radar, Camera)
Lane Marker Sensing
Obstacle
Location
DRIVE PX / DL
DRIVE PX / CUDA
MAP MODULE
Position
Sensing
GPS
-
Road Map
Local Dynamic Map
Target Path Generation
Car
Distance
Adjusting
Acceleration
Engine, Break
SPEED CONTROL
MODULE
-
Adaptive Cruise Control
Pre-Crush System
AI MODULE
-
Environment Recognition
Decision Making
Target Path Tuning
Road Structure
Road Map
Lane Distance
Steering
Adjusting
Direction
Traffic Information
-
Lane Keep Control
HMI MODULE
Big Data, Road, Traffic Information etc
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
STEERING CONTROL
MODULE
- Auto/Manual Mode SW
Operation
- System Operation Status
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
6
Autonomous Driving System Architecture
+ HMI MODULE Implementation by DRIVE CX/HMI
Driving Environment Sensing and Obstacle Recognition
-
Front Obstacles Sensing (Mili-wave Radar, Laser Radar, Camera)
Lane Marker Sensing
Obstacle
Location
DRIVE PX / DL
DRIVE PX / CUDA
MAP MODULE
Position
Sensing
GPS
-
Road Map
Local Dynamic Map
Target Path Generation
Car
Distance
Adjusting
Acceleration
Engine, Break
SPEED CONTROL
MODULE
-
Adaptive Cruise Control
Pre-Crush System
AI MODULE
-
Environment Recognition
Decision Making
Target Path Tuning
Road Structure
Road Map
Lane Distance
Steering
Adjusting
Direction
Traffic Information
-
Lane Keep Control
HMI MODULE
Big Data, Road, Traffic Information etc
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
STEERING CONTROL
MODULE
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
- Auto/Manual Mode SW
Operation
- System Operation
Status
DRIVE CX/HMI7
Autonomous Driving System Architecture
+ Computer Vision Processing by DRIVE PX/DL & CV -> Almost All Processings by Tegra
DRIVE PX / DL & CV
Computer Vision
Lane Distance
Deep Learning
VisionWorks
Obstacle
Location
DRIVE PX / DL
DRIVE PX / CUDA
MAP MODULE
Position
Sensing
GPS
-
Road Map
Local Dynamic Map
Target Path Generation
Car
Distance
SPEED CONTROL
MODULE
-
Adaptive Cruise Control
Pre-Crush System
AI MODULE
-
Environment Recognition
Decision Making
Target Path Tuning
Road Structure
Road Map
Adjusting
Acceleration
Engine, Break
Steering
Adjusting
Direction
Traffic Information
-
Lane Keep Control
HMI MODULE
Big Data, Road, Traffic Information etc
参考文献:「自動運転 システム構成と要素技術」、保坂明夫、青木啓二、津川定之
STEERING CONTROL
MODULE
森北出版、2015年7月
Reference: “Automated Driving System and Technologies”, Akio Hosaka et al,
Morikita Publishing Co., Ltd., July 2015
- Auto/Manual Mode SW
Operation
- System Operation
Status
DRIVE CX/HMI8
9
Audi zFAS Example as a Low-Speed Autonomous Driving:
Obstacle Recognition, Target Path Generation by one Tegar K1
From
GTC2015
10
Deep Learning Revolutionize Computer Vision
Required Separate Algorithms/Apps
- Pedestrian: HOG etc
- Traffic Sign: Hough Transform + Character Recog. etc
Only simple context recognition
- Pedestrian Y/N Only (no additional info)
- Speed Limit Signs Only
One Deep Neural Net App can Detect various Objects
- Pedestrian, Cars, Traffic Signs, lanes
- Also with many attributes (Car: Police Car, Van, Sedan, Truck, Ambulance….)
DEEP NEURAL NETWORK
CONVENTIONAL
(…)
11
TEGRA X1
CLASSIFICATION Performance
IMAGES / SECOND
AlexNet
100
90
80
70
60
50
40
30
20
10
0
Tegra K1
Tegra X1
12
13
Growing Performance of Automotive Tegra Products
will allow further Integration in the Future
Tegra X1
1200
Tegra X1 (FP16)
Core i7
1000
GPU
GPU
CPU
800
CPU
GFLOPS
FP16/INT16
600
400
Tegra K1
200
Tegra 4
Tegra 2
Tegra 3
0
TIME
Note: 4790K Core i7, CPU @ 4GHz,14
GPU @ 350 MHz
DRIVE PX
For Map Module, AI Module and Computer Vision
15
An advanced computing platform based on NVIDIA
Tegra processors for autonomous driving cars
FEATURES
DRIVE PX
The ability to capture and process multiple
HD camera and sensor inputs
A rich middleware for computer graphics,
computer vision and deep learning
A powerful and easy to develop platform for
algorithm research and rapid prototyping
NVIDIA CONFIDENTIAL — DRIVE PX DEVELOPMENT PLATFORM
16
Preliminary information — Subject to change
DRIVE PX Camera & Display Interfaces
 12 simultaneous LVDS
camera inputs
• All cameras
synchronized within
each Group (3 groups)
 2 LVDS display ports
Group A
Group B
Group C
Display
17 & Confidential
Proprietary
All Information Subject to Change
Other Interfaces to Aurix
CAN*, LIN*, FlexRay* and Ethernet
Ethernet (x1)
1x Power
UART (x1)
FlexRay (x2)
LIN (x4)
48-pin Automotive Grade
Vehicle Harness
CAN 2.0 (x6)
18
Dual Tegra X1 VCM; each VCM consists of:
Tegra X1 processor
DRAM: 4GB
NOR FLASH: 64MB
Hardware Specs
PROCESSORS
eMMC: 64GB
Inter-Tegra X1 VCM Communication
SPI and USB 3.0 for direct inter-Tegra communication and through
Ethernet Switch
ASIL-D MCU
Camera and IO controls through ASIL-D MCU.
NVIDIA CONFIDENTIAL — DRIVE PX DEVELOPMENT PLATFORM
19
Preliminary information — Subject to change
Sensors:
Vision Sensors interface:
12x LVDS Cameras
Hardware Specs
PERIPHERALS
Sensor Interfaces for Radar, LIDAR, Vehicle Dynamics etc.:
CAN 2.0; LIN; Ethernet; Flexray
Displays:
LVDS interface (x2)
Power Management of ECU:
System power monitor/control — ASIL-D MCU
NVIDIA CONFIDENTIAL — DRIVE PX DEVELOPMENT PLATFORM
20
Preliminary information — Subject to change
OS: NVIDIA Vibrante Linux 4.0
64-bit Kernel Linux, Quickboot, AutoSAR RunTimeEnvironment
Graphics:
Open GL ES 3.1
Development Tools/Samples: Delivered through Jetpack 2.x
DRIVE PX
software specs
Graphics debugger, PerfKit, DNN Classifier Sample,
Vision Works 1.0 (beta) Computer Vision libraries and Samples
ASIL MCU Support for CAN, Ethernet, Flexray and LIN; AutoSAR framework
External Storage for Video Recording
USB3.0 interface for camera output in RAW or H.265/H.264 encoded formats
Camera: NVMedia and Driver support for LVDS camera
Open Source Collaboration initiatives/Compliance:
Yocto 1.8
Genivi7 Compliant
NVIDIA CONFIDENTIAL — DRIVE PX DEVELOPMENT PLATFORM
21
Preliminary information — Subject to change
WORLD CLASS SOFTWARE TOOLS
Faster debug and analysis reduces development costs
TEGRA GRAPHICS DEBUGGER
PERFKIT
ECLIPSE IDE
Visualize GPU performance metrics
Performance monitoring
Standard Linux development environment
Automated analysis of GPU bottlenecks
Automated bottleneck analysis
22
Preliminary information — Subject to change
DRIVE PX LINUX SOFTWARE STACK
Performance Microprocessor A
Performance Microprocessor B
Safety MCU
Applications
Applications
Applications
Graphics/Compute
Graphics/Compute
CV/DL Libraries
NVMedia
Filesystem(s)
Linux
BSP/Drivers
Imaging
(Camera)
Pipeline
Graphics/Compute
AUTOSAR
BSW on
Linux
AUTOSAR
on
Safety
MCU
CUDA/EGL/
Open GL ES
AUTOSAR
BSW on
Linux
MCA
L
CV/DL Libraries
Graphics/Compute
CUDA/EGL/
Open GL ES
Linux
Imaging
(Camera)
Pipeline
Safety MCU
OS/3rd SW/HW
Elektrobit
NVIDIA Licensed SW
Filesystem(s)
Linux
BSP/Drivers
Linux
Tegra™ X1 Hardware (ARM, GPU & SoC Peripherals)
T1/OEM SW
NVMedia
Tegra™ X1 Hardware (ARM, GPU & SoC Peripherals)
Drive PX Hardware
23
Preliminary information — Subject to change
DRIVE CX
For HMI MODULE
24
THE SOUL OF
NVIDIA DRIVE™ CX
DIGITAL COCKPIT CAR COMPUTER
Natural Speech
OTA updates
Advanced Visuals
Hypervisor – Cluster Cockpit
NVIDIA CONFIDENTIAL
25
ADVANCED VISUALS – Digital CLUSTER
TODAY
DRIVE CX
26
DRIVE CX ADAS
Also supported by DRIVE PX
Best-in-Class
Surround View
27
NVIDIA DRIVE Design
NVIDIA’s HMI Platform version 8.0
Design Studio
Professional artist environment
Design Architect
Integrated engineering
environment
28
Fail-safe NATURAL LANGUAGE SPEECH
ACCURACY
VOCABULARY
SPEED
Today
(no internet
connection)
Google
(with Internet
Connection)
DRIVE CX
(no internet
connection)
LOW
HIGH
HIGH
1M parameters
30M parameters
30M parameters
SMALL
LARGE
LARGE
50k words
4M words
4M words
FAST
… or no response
(lost internet
connection)
FAST
… always
SLOW
500+ ms latency
29
SUMMARY
1. Autonomous Driving System Architecture consists of Sensing Module, Map
Module, AI Module and HMI Module. DRIVE PX and CX can implement all
functions with CUDA, Deep Learning , Computer Vision and HMI Frameworks.
2. DRIVE PX consists of two powerful Tegra X1 processors with the total
performance of 2.3TFLOPS. It comes with a rich middleware for GPU
Computing, Deep Learning and Computer Vision.
3. DRIVE CX powerful Tegra X1 processor enables the fail-safe Natural Speech
Recognition, advanced visual quality which offers a safe, versatile and highquality HMI. This is essential for the critical human-car interaction in the
Autonomous Driving Cars.
4. Today, we might start with a few DRIVE PX and a DRIVE CX. However, the
continuous performance and feature enhancement in the future will make it
possible to implement the total system by a single DRIVE platform if required.
30
THANK YOU