Mike Dempsey Managing Director - Microsoft · –For example: Ibeo Scala B2, Velodyne VLP-16 “Puck”, Delphi ESR 2.5, Continental ARS430 –Produce the same output message format

Copyright © Claytex Services Limited 2018

Virtual development and testing

of autonomous vehicles

Mike Dempsey

Managing Director


Who are Claytex?

• Model-based engineering analysis consultancy

– Innovators in CAE process

– Leading the way on zero-prototype development

– Specialists in high-fidelity real-time simulation

– Users of Dymola and Modelica since 1999

• Provider of software solutions for systems engineering

– Dymola distributors since 2003

– Dassault Systemes partner since 2008

– rFpro system integrator and distributor since 2009

• Modelica library and FMI tool developers

• Dassault Systemes Certified Education Partner

• Offices in the UK, USA and South Africa

• Major customers include Automotive OEM’s, suppliers and

Motorsport teams (Formula 1, NASCAR, Indycar, Formula E)


Autonomous Vehicles Are Coming!


How do we prove the AI is safe?

• Research has been carried out to

determine the amount of testing

required to prove autonomous

vehicles are safer than human

drivers

• Each software/hardware release

will have to be validated

• However, this is not physically

possible

Source: Rand Corporation


Testing autonomous vehicles

Simulation

• Huge number of scenarios can be considered

• Full control of virtual environment: traffic, pedestrians, weather, etc.

Proving Grounds

• Recreate critical scenarios

• Limited control of the environment

• Robot controlled targets

• Pedestrian targets

• No control of weather and light

Field Tests

• Investigation of real driving situations

• No control of the environment


What must our simulation tool do to support this?

• Provide accurate models of real world locations

• Support interaction with traffic, pedestrians, animals

• Have a flexible approach to defining test scenarios

– Weather, time of day, location, etc.

• Allow human controlled vehicles to interact in the environment

• Allow multiple CAV to operate in the same environment

• Run in real-time to support HiL integration of the AI processor

• Provide sensor models that replicate the real sensors

– radar, LiDAR, ultrasound, camera, GPS, etc.

• Support the integration of your vehicle physics model



Real world locations

• 100km’s of real world locations are available

• Proving grounds: Idiada, Millbrook, Mcity, SMLL

• Race tracks: Nordschleife, full Formula 1 and NASCAR

calendars, partial Formula E calendar and many other

tracks worldwide

• Public roads including locations in:

– France, Germany, USA, Canada, Italy, China


Traffic

• rFpro connects to wide range of traffic

modelling tools

– SUMO, IPG Traffic, PTV Vissim, Vires VTD,

Forum 8

• The traffic pack of vehicles covers road cars,

HGV and pedestrians

• Allows control of traffic lights, etc.


Control weather conditions


Control the lighting conditions


Wet roads


Vehicle Physics

• Wraps around your vehicle model

– Supports Dymola, CarMaker, CarSim,

SIMPACK, Dynaware, AVL_VSM, VI-

Grade, dSPACE_ASM, Simulink and C++

• Includes the rFpro TERRAIN_SERVER

– Provides high frequency road surface

information to the vehicle model

• Supports integration with HIL

– Off-the-shelf implementations for

Concurrent Realtime, dSPACE,

Speedgoat

• Claytex uses Dymola with multi-domain

vehicle models


Vehicle Systems Modelling and Analysis

• Suite of Modelica libraries for Vehicle

Systems Modelling and Analysis

• Core platform enables performance,

fuel economy and energy analysis

– Drive cycle simulation

• Application specific extensions

provide detailed models across many

areas

– Engines, powertrain, vehicle

dynamics, driver-in-the-loop

• Used for modelling conventional,

hybrid and electric vehicles

• Real-time capable

• Integrates with rFpro


Testing entire toolchain

rFpro

Sensor model

Algorithms

Validate vs Ground truth

Vehicle physics

Control systems

HUMAN TEST

DRIVER


Sensor feeds

• Apply effects to replicate what the

sensors see

– Lens distortion effects

– Dirt

• Normal and Fisheye camera views

• Depth map and object lists

• Open API to access all this data


Sensor modelling

• Building a library of models that represent real

sensors

– For example: Ibeo Scala B2, Velodyne VLP-16 “Puck”,

Delphi ESR 2.5, Continental ARS430

– Produce the same output message format

– Capture the dynamics of the sensor e.g. rotation of

sensor with time

• Developing control panel to manage and configure

scenarios, sensor setup, etc.

• Active R&D projects to increase fidelity of sensor

models under 2 Innovate UK projects

– Sensor models will take into account weather effects and

other significant sources of noise



dRISK – AV risk assessment


Summary

• Simulation will be an essential part of the

development and testing of autonomous

vehicles

• Your tools need to provide:

– detailed models of the environment

– support for HIL to include the real controller

– representative sensor models

– vehicle physics model

• rFpro offers all these capabilities

Documents

Mike Dempsey Managing Director - Microsoft · –For example: Ibeo Scala B2, Velodyne VLP-16 “Puck”, Delphi ESR 2.5, Continental ARS430 –Produce the same output message format