ACCELERATING THE RACE TO SELF-DRIVING CARS. Jen-Hsun Huang, Co-Founder & CEO, NVIDIA Jan. 4, 2016

Size: px

Start display at page:

Download "ACCELERATING THE RACE TO SELF-DRIVING CARS. Jen-Hsun Huang, Co-Founder & CEO, NVIDIA Jan. 4, 2016"

Octavia Garrison
6 years ago
Views:

1 ACCELERATING THE RACE TO SELF-DRIVING CARS Jen-Hsun Huang, Co-Founder & CEO, NVIDIA Jan. 4, 2016

2 SELF-DRIVING IS A MAJOR COMPUTER SCIENCE CHALLENGE SOFTWARE SUPERCOMPUTER DEEP LEARNING 2

3 NVIDIA DRIVE PX 2 12 CPU cores Pascal GPU 8 TFLOPS 24 DL TOPS 16nm FF 250W Liquid Cooled World s First AI Supercomputer for Self-Driving Cars 3

4 THE SELF-DRIVING REVOLUTION Safer Driving New Mobility Services Urban Redesign 4

5 TWO VISIONS 5

6 THE BASIC SELF-DRIVING LOOP MAP SENSE CONTROL LOCALIZE PLAN PERCEIVE 6

7 SELF-DRIVING IS HARD The world is complex The world is unpredictable The world is hazardous 7

8 DEEP LEARNING TO THE RESCUE DNN BIG DATA GPU The GPU is the workhorse of modern A.I. 8

9 THE AI RACE IS ON 100% 90% Traditional CV IMAGENET Accuracy Rate Deep Learning 80% 70% 60% IBM Watson Achieves Breakthrough in Natural Language Processing Facebook Launches Big Sur Baidu Deep Speech 2 Beats Humans 50% 40% 30% 20% 10% 0% Google Launches TensorFlow Toyota Invests $1B in AI Labs Microsoft & U. Science & Tech, China Beat Humans on IQ 9

10 THE ENGINE OF MODERN AI EDUCATION BIG SUR TENSORFLOW WATSON CNTK TORCH CAFFE THEANO MATCONVNET MOCHA.JL PURINE START-UPS CHAINER DL4J KERAS OPENDEEP MINERVA MXNET* SCHULTS LABORATORIES VITRUVIAN NVIDIA GPU PLATFORM *U. Washington, CMU, Stanford, TuSimple, NYU, Microsoft, U. Alberta, MIT, NYU Shanghai 10

11 DEEP LEARNING EVERYWHERE NVIDIA Jetson NVIDIA Tesla NVIDIA DRIVE PX NVIDIA Titan X 11

12 END-TO-END DEEP LEARNING PLATFORM FOR SELF-DRIVING CARS DNN TRAINING PLATFORM IN-CAR AI SUPERCOMPUTER DEEP NEURAL NETWORK 12

operations Single and multi-class detection Segmentation 70% 60%

13 100% KITTI Dataset: Object Detection NVIDIA DRIVENet 90% 80% Top Score 88% 9 inception layers 3 convolutional layers 37M neurons 40B operations Single and multi-class detection Segmentation 70% 60% 50% 40% 39% 55% 72% 30% 7/2015 8/2015 9/ / / /

14 KITTI dataset 14

15 Courtesy of Cityscapes dataset 15 project

16 Courtesy of Cityscapes dataset 16 project

17 Courtesy 17 of Audi

18 Using NVIDIA DIGITS deep learning platform, in less than four hours we achieved over 96% accuracy using Ruhr University Bochum s traffic sign database. While others invested years of development to achieve similar levels of perception with classical computer vision algorithms, we have been able to do it at the speed of light. Matthias Rudolph, Director of Architecture, Driver Assistance Systems, Audi 18

and exceeded the performance of classic computer vision.

19 Due to deep learning, we brought the vehicle s environment perception a significant step closer to human performance and exceeded the performance of classic computer vision. Ralf G. Herrtwich Director of Vehicle Automation, Daimler 19

20 Deep learning technology can dramatically improve accuracy of detection and decisionmaking algorithms for autonomous driving. ZMP is achieving remarkable results using deep neural networks on NVIDIA GPUs for pedestrian detection. We will expand our use of deep learning on NVIDIA GPUs to realize our driverless Robot Taxi service. Image from ZMP [Sahin] Hisashi Taniguchi CEO, ZMP Inc. 20

BMW is exploring the use of deep learning for

autonomous driving to quality inspection in

The ability to rapidly train deep neural

21 BMW is exploring the use of deep learning for a wide range of automotive use cases, from autonomous driving to quality inspection in manufacturing. The ability to rapidly train deep neural networks on vast amounts of data is critical. Using an NVIDIA GPU cluster with NVIDIA DIGITS, we are achieving excellent results. Uwe Higgen Head of BMW Group Technology Office (USA) 21

With the NVIDIA deep learning platform, we have greatly improved performance on a variety of image recognition applications for cars, surveillance cameras and robotics.

22 With the NVIDIA deep learning platform, we have greatly improved performance on a variety of image recognition applications for cars, surveillance cameras and robotics. The remarkable thing is that we did it all with a single NVIDIA GPU-powered deep neural network, in a very short time. Daisuke Okanohara Founder & SVP, Preferred Networks Distributed Cooperative Deep Learning 22

tested and developed as we move them onto the NVIDIA DRIVE PX.

23 Deep learning on NVIDIA DIGITS has allowed for a 30x enhancement in training pedestrian detection algorithms, which are being further tested and developed as we move them onto the NVIDIA DRIVE PX. Dragos Maciuca, Technical Director, Ford Research and Innovation Center 23

24 MANY THINGS TO LEARN 24

25 END-TO-END DEEP LEARNING PLATFORM FOR SELF-DRIVING CARS DRIVEWORKS Perception Localization Planning Visualization NVIDIA DIGITS NVIDIA DRIVE PX 2 NVIDIA DRIVENET 25

26 26

27 27

28 NVIDIA DRIVE PX 2 TITAN X DRIVE PX 2 Process 28nm 16nm FinFET CPU 12 CPU cores 8-core A core Denver GPU Maxwell Pascal TFLOPS 7 8 DL TOPS 7 24 AlexNet 450 images / sec 2,800 images / sec 28

29 150 MACBOOK PROS IN YOUR TRUNK 6 TITAN X = 42 TFLOPS, Core i7 = 280 GFLOPS, 42 / 0.28 = 150 MacBook Pros 29

30 2 NEXT-GEN TEGRA PROCESSORS 30

31 2 NEXT-GEN PASCAL GPUS 31

32 LIQUID COOLING 250W Operation up to 80C (175F) ambient 256 cubic inches 32

33 NVIDIA DRIVE PX 2 33

34 NVIDIA s Deep Learning Car Computer Selected by Volvo on Journey Towards a Crash-Free Future 34

35 NVIDIA DRIVE PX SELF-DRIVING CAR PLATFORM DRIVEWORKS Perception Localization Planning Visualization NVIDIA DIGITS NVIDIA DRIVE PX 2 NVIDIA DRIVENET 35

FUNCTIONAL SAFETY FOR AUTONOMOUS DRIVING

FUNCTIONAL SAFETY FOR AUTONOMOUS DRIVING Dr. Justyna Zander, NVIDIA January 30, 2017 IS&T Int. Symposium on Electronic Imaging 2017; Autonomous Vehicles and Machines 2017; 29 January - 2 February, 2017