Telematics Wire Magazine 2018

Page 14

THOUGHT LEADERS

The Role of Simulation in Development of Reliable and Safe Autonomous Vehicle manufacturing and will change the society forever. But crea ng autonomous vehicle is a formidable technological challenge. The autonomous vehicle can be described as a super computer crunching tens of terabytes of data during a few hours of driving. Developing autonomous vehicle technology requires ambi ous new development in various technologies such as sensing, machine learning and ar ficial intelligence.

Ashok Khondge Principal Engineer, ANSYS Customer Excellence, ANSYS, Inc. Ashok Khondge serves as automo ve CAE applica on specialist for Asia region and part of ANSYS Customer Excellence team. Over past 15 years he has held product, applica ons and leadership roles providing CAE solu ons to Automo ve industry.

Leading automakers across the world are giving high priority to development of autonomous vehicles. It is a formidable technological challenge. Studies show that billions of miles of road testing will be necessary to ensure safety and reliability of autonomous vehicles, yet time-tomarket is short with competition heating up. This seemingly difficult task can only be accomplished using simulation with precision, speed and within available resources.

Background and Challenges Connected and smart vehicles, Ride sharing, Electrifica on and Autonomous vehicles are disrup ng transporta on industry today. The global compe on to develop such technologies is hea ng up. New business models will evolve taking advantage of such technology innova ons. The first company to have the upper hand will win and sustain. The autonomous vehicle revolu on will have a major impact on the vehicle PG.14 | Smart Automo ve | Mar - Apr 2018

The reliable development of machine learning based vision and percep on models that imitate human driver under all possible driving condi on is a major problem. An autonomous vehicle’s computer needs a 360-degree surround view. It must recognize other vehicles, road signs, pedestrians, markings, trees, buildings, traffic lights, and several other things under all possible driving scenarios and weather condi ons such as in the darkness of night or while fogging, in rain and in snow etc. This is a difficult problem to solve using rule based computer algorithms. Autonomous vehicle engineers are using machine learning based neural network methods that can be trained using data rather than programmed. Autonomous vehicle engineers train such computer models by feeding them with enough test data to sufficiently react to mul ple scenarios. However, the problem is, it is not easy or safe to replicate many scenarios in the real-world environments. A report by RAND [1] indicate that autonomous vehicles would have to be driven hundreds of millions of miles or some mes billions of miles to demonstrate acceptable reliability. Under very aggressive tes ng assump ons, it would take tens of years or may be hundreds of years to drive these many miles. This is an impossible proposi on to develop reliable and safe autonomous vehicles.

and Ola, opera ng within certain city limit with constrained border lowering the technological barrier. However, this may s ll take several years to develop autonomous vehicles given that machine learning based neural network algorithms need a lot of data to make them reliable, and safe. The other op on is to build machine learning models that will be able to train with li le data. However, such technology development is in early phase of research. Finally, it is the simula on, if build properly can help autonomous vehicle engineers to gather enough training data and test their algorithms in quick me. Simula on has been used in automo ve industry for several decades with proven record of accelera ng technology development. With simula on, thousands of virtual tests can be performed on virtual prototypes using computers, enabling accelera on in technology development within a frac on of budget and me required for physical tes ng. Simula on provides three broad benefits namely – Faster me-to-market, Reduced cost, and Enhanced product quality. Autonomous vehicle system is essen ally a control loop, comprising of four important elements – physical world, sensors, controllers and actuators (Figure 1). An autonomous drive control so ware is embedded on controller with ability to train itself while performing virtual tests such as Model in loop (MiL), So ware in Loop (SiL), and Hardware in Loop (HiL). To replicate this autonomous driving system

Fast-tracking Reliable and Safe Autonomy To fast track the development of autonomous vehicles, auto OEMs may start rolling out autonomous vehicles suitable for constrained environment. OEMs may collaborate with likes of Uber

Figure-1 Essential elements of autonomous system control

www.telema cswire.net I www.coe-iot.com


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.