A workshop is planned for December 10th in Palo Alto to plan the work to be done for the AVP demonstration and design the architecture.
In the week prior to the demonstration, a hackathon will be held at Apex.AI's office in Palo Alto, California. The purpose of the hackathon will be to polish the demonstration.
The hackathon will be held from April 20th to April 23rd.
Discussion on the AVP demo
Discussion over whether the Autoware.AI release dates and the hackathon clash, and if we should move the hackathon to after the Autoware.AI release.
It was decided not to change the dates as enough resources are expected to be available on both projects.
Current planned attendees for the planning workshop are:
LGE (Brian or Dmitry)
AutonomouStuff (Josh Whitley)
Apex.AI will produce a detailed architecture sketch and circulate this prior to the planning workshop.
Embotech is having trouble with integration of Autoware.Auto with LGSVL. LGE will help out.
Still missing pieces of the complete self-driving stack:
Integration (launch files, parameters, etc.) and automated/manual testing: Someone needs to write ROS 2 launch files for integrating the different parts of the pipeline and for the AVP demos. Someone needs to set up parameter configuration files correct for the hardware and application. Someone needs to write integration tests using the launch_testing framework.
Pose fusion (odometry and NDT): Someone needs to write a fusor node, e.g. an EKF pose fusor, either by porting the one from Autoware.AI or writing a new one. The fusion will not include the IMU data, just the odometry data and NDT output.
Tools and GUIs: Collecting rosbags, an easy way to store and download rosbags for debugging, an rviz setup for visualisation, etc.
Safety aspects (functional and operational): We need a safety plan and, hopefully, a safety case for the demonstration.
Since we will be targetting Eloquent for the demo, we need to:
Port Autoware.Auto to Eloquent
Get CI going for Eloquent
Decide if we should use XML or Python for our launch files; the XML launch files are quite limited in functionality compared to Python launch files
Silexica may be interested in taking on the integration task as it might fit with their goals for extending their own tool's demo. They will make a decision in the next few days.
The Apex.AI car (Lexus SUV) has two Velodynes, a very expensive GPS, and an Xsense IMU. They already have a URDF description of the car and sensors and a mesh for simulation/visualisation. It uses an automatic transmission and has a drive-by-wire interface from AutonomouStuff. It can autonomously shift between drive and reverse.
Apex.AI (Dejan) and AutonomouStuff (Josh) to find out which ROS drivers for sensors are available, then gather data in the AutonomouStuff carpark using them.
Apex.AI (Dejan) to get the physical properties of the car, or AutonomouStuff (Josh) if they cannot.
Parkopedia is best positioned to provide the safety case.
We need to define in detail the AVP application requirements, and how we will specify where to drop off the car, where to park, etc.
Parkopedia already has an app (based on ROS 1) for controlling the AVP application, which they may be able to offer for controlling the demo.
The German automobile manufacturers association has defined five phases through which AVP proceeds.
The ISO standard for AVP that is under preparation is focused on the relationship between the various actors in the AVP application, such as the vehicle and the traffic management system, to manage things like knowing if a vehicle can fit in a carpark, if there is a space available, where it is, etc.