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

Autoware Simulation WG: Meeting Minutes 21 Aug 2019

The Autoware Simulation Working Group Meeting, held on 21 August 2019.
In this meeting, we discuss about the requirements of the simulation software for developing Autoware.

Attendant

  • Masaya Kataoka (TierIV)
  • Kenji Miyake (TierIV)
  • Geoffrey Biggs (Tier IV)
  • Brian Shin (LGSVL)
  • Dmitry Zelenkovsky (LGSVL)
  • Ian Colwell (Autonomous Staff)
  • Misha Savchenko (metamoto)

We talk about some features which is very important for Autonomous Driving simulation.

  1. Determinism
  • LGSVL
    New version of LGSVL simulator includes determinism in physical point of view, however, the sensor/rendering point of view, it does not works deterministic.

  • metamoto
    metamoto simulation platform is based on Unity, but they render without using Unity Pipeline.
    So, it works determionistic.

  1. Scenario Import/Export
    We want to use rosbag in order to import Log data from the real vehicle.
    We want to use OpenScenario in order to use multiple simulator easilly.
  • LGSVL
    LGSVL will supports importing rosbag data from real vehicle.
    Now, LGSVL publish ground truth data via ROS topic and Python API return.
    So, we can generate scenario in LGSVL simulator by using this rosbag file.
    Now, openscenario format is not supported.
  • metamoto
    Now, OpenScenario does not supported, but it will support in the future.
  1. Map Editing/Format
    We have to edit map when we try to make new simulation environment.
  • LGSVL
    We can generate Simulation environment by using VectorZero RoadRunner.
    Forsight.ai can provide HD map.
    parallel domain also can provide HD map.

  • metamoto
    We can generate Simulation environment by using VectorZero RoadRunner.

  1. Parameter Sweep
  • LGSVL
    Now, there is only NPC random seed feature exist.
    Ego vehicle parameter sweep does not supported yet.

  • metamoto
    metamoto has a parameter sweep feature.

  1. Photo Realism/Dataset Collection/Sensor Simulation
  • LGSVL
    LGSVL has a new rendering pipeline and data collection Python API.
    Whether simulation does not works in a Lidar simulation.
  • metamoto
    Camera simulation is Photo-realistic, metamoto can simulate snow, rain, fog in camera simulation and simulate rain in Lidar simulation.
  1. Vehicle Configuration/ URDF import
  • LGSVL
    Now, udrf import does not support.
  • metamoto
    Now, udrf import does not support.
    Only GUI web tools can edit vehicle configuration.
1 Like

That’s great!
Can you share the meeting recording?

Alon.