Accele lerating AV Productization wit ith AI Danny Atsmon - CEO, - - PowerPoint PPT Presentation

Accele lerating AV Productization wit ith AI

Danny Atsmon - CEO, Cognata Simon Berard – CATIA Strategy Senior Manager, Dassault Systemes

Agenda

• Vision
• Challenges

○

AI Solutions in Design

○

AI Solutions in Validation

• Solution Convergence

AI AI-driven Desig ign and Vali lidation Solu lutions

Vis ision

Design Challenges

3x more System

Requirements

Chall llenge: Mis issio ion Driv iven Engin ineerin ing

Autonomous Systems Traditional Vehicles Taxi Driver Delivery People Mission lef eft t to

• the Use

User Mission en engineered within the Sys yste tem

Costs vs Comple lexity – AI I is is makin ing it it possible

Functional Requirements

AUTONOMOUS VEHICLES

Leverage Patrimony MASS- CUSTOMIZE ZED CARS RS

Explore Variants MOBILITY EXPERIENCE

Business Driven Innovation COMPONENTS

Structure Legacy

AI I is is brin ingin ing Quali lity

Functional Requirements

Chall llenge 2: : Physic ically Exa xact

Multidiscipline, Multiphysics, Multiscale Consistent System Experience Validation Sensors optimization

AI I Solu lutions fo for Design

Learning from Patrimony Context Sensitive Automated Assembly Parameters space Exploration Function Driven Generative Design Model Based System Engineering Performance Tradeoff

Valid lidation Challenges

AV Tech Has Had Some Unpla lanned Setb tbacks

Chall llenge #1: Scale

Functional Requirements Level of Autonomy

L0: INFO FORM

__________

Blind Spot Lane Warning Park Assist L1: ASSIST

__________

Adaptive Cruise Control L2: ASSIST

__________

Adaptive Cruise Control + Lane Centering L2+: ASSIST

__________

Highway Chauffeur L4: DR DRIVE

__________

Driverless Taxi

Chall llenge #1: Scale - Need Vs. . Actu tual

The Need: 11 11B* B*

*Rand corporation

Actually dri riven: 16M (0 (0.15%)

Chall llenge #2: Realis ism

REAL LIF IFE SIM IMULATION

VS.

Realism

100%

We need a reali listic sim imulation fo for a meaningful l coverage

Scalable Realism

??%

Scalable

Reali lism - Uncanny Vall lley (M (Masahiro Mori, i, 1978)

16

Completely machine like EMOTIONAL RESPONSE +

• Familiarity

50% 100%

UNCANNY VALLEY

Human likeness

Moving Still Fully human Industrial robot Stuffed animal Bunarkupuppet Polar Express Humanoid robot Zombie Prosthetic hand Simulation today

Reali lism - Uncanny Vall lley

17

Completely machine like EMOTIONAL RESPONSE +

• Familiarity

50 50% 100%

UNCANNY VALLEY

Human likeness

Moving Still Fully human Industrial robot Stuffed animal Bunarkupuppet Polar Express Humanoid robot Zombie Prosthetic hand Simulation needs to be here

The anim imation solu lution

18

• Walt Disney

“An animator cannot capture all

• f reality . Instead he picks 3 or

4 distinct elements and exaggerate them.“

AI I based meth thods fo for Realis ism

Nvidia & MIT - Video (Labels) to Video Synthesis – Wang et. al. 2018

END TO END

Procedural Modeling of a Building from a Single Image – Nishida et al, 2018

LAYERED APPROACH LAYERED APPROACH

Learning from Synthetic Humans – Varol et. al 2017

Realistic, not consistent Consistent, Not scalable (Manual) Consistent, Not scalable (Variations)

Chall llenge #2: : Realis ism conclu lusio ions

REAL LIFE SIMULATION

VS.

• DNN transfer functions ~= Realism
• Isolated layers brings better results than End to End
• Data sources should be wide (crowd sourced)

Cognata - 4 4 Technolo logy lay layers

21

STATIC DYNAMIC SENSING CLOUD & ANALYTICS

A Combined Solution

Desig ign and Vali lidation Converge

Better Tog

• gether
• Tests and designs together
• Validation base directly from

requirements

• Smart coverage

Takeaways

• Autonomous vehicles brings

○ New designs and use cases ○ Large scale validation challenge

• AI is a key to get Autonomous vehicles in a safe and cost effective way
• A platform solution to manage, design and validate is the needed solution

Thank You!

End to to end li limitations

26

The Problem: Not consistent, overfeat

Buil ildin ings reconstruction conclusions

27

The Good: Consistent, Procedural The Bad: Not practical

Learning pose - Conclu lusio ions

28

This the most advanced way of learning moving objects.

The Good: Consistent The Bad: Not scalable