ROS Resources: Documentation | Support | Discussion Forum | Service Status | Q&A answers.ros.org

Report on the Autoware workshop at IV 2019

Autoware Tutorial @IV 2019 (9 June)

What is Autoware?

Autoware is one of the 2 (the other being Apollo) open source software stacks for autonomous driving.

What is IV 2019?

Intelligent Vehicles is the number one annual conference for industry and academia doing research and development of autonomous vehicles.

What is Autoware Tutorial@IV2019?

This is the first-of-a-kind event that was organized at IV2019 on June 9.
The event consisted of presentations and tutorials on features in Autoware as well as applications that were built with Autoware.

Around 100 people participated and saw 20 presentations/tutorials: https://www.autoware.org/agenda.

Short minutes of the most important presented information are listed below.

3D perception in Autoware.Auto | Christopher Ho (Apex.AI)

Chris’ talk was very well received and it was referenced throughout the whole day as a path to production.

Computer Vision | Jacob Lambert (Nagoya University & Perception Engine)

Jacob presented an interesting study comparing 9 different LiDAR sensors measuring different environments at different times of the day.
According to him Velodyne and Hesai perform the best.

Coupling Autoware and Simcenter Prescan for virtual testing and verification of automated driving systems | Frank Rijks (Siemens PLM Software)

Siemens presented two things:

  1. Their PreScan simulator, which is a physics-based simulation for AD and ADAS systems.
    Among other things they know how to simulate radar sensors

  2. A full demo in which they had Autoware drive their vehicle on a public road in the Netherlands

Model-based Systems Engineering | Bernd Gassman & Frederik Pasch (Intel Germany)

Intel presented an model-based approach in which they abstract away ROS 1 constructs and hide them behind a configuration GUI from which they generate code stubs.
They abstract way different types of communication (asynchronous, synchronous), ROS 1 nodes, publishers/subscribers, the parameter server and also add a life cycle to ROS 1 nodes.
Then they generate an implementation file in which the user is expected to fill-in the application-specific code.

This is not yet open-source but Intel would consider making it open-source if they get lots of demand.

Responsibility-Sensitive Safety | Bernd Gassman (Intel Germany)

Intel also presented an RSS paper from MobilEye.

The implementation is available here.

At its core, the RSS model is designed to formalize and contextualize human judgment regarding all driving situations and dilemmas.
This formalization is then translated into a set of rules in four essential realms.
First, RSS defines safe distances for all driving scenarios, from one-way traffic to junctions and multiple geometry scenarios.
Secondly, RSS defines exactly what is considered to be a dangerous situation, as a derivative of all the semantics and rules defined in the “human judgment” formalization.
The third essential definition in the model is the proper response that needs to be taken in order to evade a dangerous situation.
By formalizing those first three realms, we create a set of parameters that are lacking in today’s development of autonomous vehicle safety systems.

Maps for Autonomous Parking | Angelo Mastroberardino, Punnu Phairatt, Brian Holt (Parkopedia)

3D Mapping for Autoware | Simon Thompson (TierIV)

There were 2 talks on how to integrate an increasingly popular Lanelet2 format/library into Autoware.
Lanelet2 is a format/library for creating and serving maps for autonomous driving.

Adopting Manycore as Autoware Accelerator | Stephane Strahm (Kalray)

Kalray presented the MPPA (Massively Parallel Processors Array), which is a manycore architecture for compute and CV/CNN acceleration.
As an accelerator it interfaces with the x86 or ARM architecture.
MPPA itself runs an in-house operating system, Linux, and third-party operating systems such as eMCOS from eSol.
They showed how you can port ndt_matching and Yolov3 CNN-based vision algorithm to MPPA and they do that using common tools (gcc, libc++, openCL, …).

Lessons learned: Integration of Autoware at our demonstrator | Markus Schratter, Konstantin Lassnig, Daniel Watzenig (Virtual Vehicle Research Center)

Researchers from TU Graz showed how they use Autoware to drive autonomously around their university campus using a Ford vehicle.

CARLA: Open-source Simulator for Autonomous Driving Research | Nestor Subiron

(Computer Vision Center)

Integration of Carla with Autoware | Frederik Pasch (Intel Germany)

Intel and CVC presented Carla (an Unreal-based autonomous driving simulator) and how to integrate it into Autoware.

One challenge that was noticed is that Carla decided to use OpenDrive as a format for AD maps while in Autoware we are now transitioning to Lanelet2.

1 Like