Lidar sensor possibly wrongly modeled in carla

[anshulpaigwar]

While using Carla 0.91 (and also in previous versions) I observed that there are always some dead angles in the Lidar data as shown in the image below from pygame window (manual_control.py):

Others have also noticed the problem as in #995.

I tried to configure the settings parameters of Lidar sensor but could not resolve it. But I noticed that keeping the rotation frequency constant (10Hz) if I increased points_per_second the dead angle reduces while it is more if the points_per_second are low.
Also dead angles varies on different system with same configuration.

Upon checking the code RayCastLidar.cpp:

  const float CurrentHorizontalAngle = LidarMeasurement.GetHorizontalAngle();
  const float AngleDistanceOfTick = Description.RotationFrequency * 360.0f * DeltaTime;
  const float AngleDistanceOfLaserMeasure = AngleDistanceOfTick / PointsToScanWithOneLaser;

  LidarMeasurement.Reset(ChannelCount * PointsToScanWithOneLaser);

  for (auto Channel = 0u; Channel < ChannelCount; ++Channel)
  {
    for (auto i = 0u; i < PointsToScanWithOneLaser; ++i)
    {
      FVector Point;
      const float Angle = CurrentHorizontalAngle + AngleDistanceOfLaserMeasure * i;
      if (ShootLaser(Channel, Angle, Point))
      {
        LidarMeasurement.WritePoint(Channel, Point);
      }
    }
}

The for loop is from 0 to PointsToScanWithOneLaser which explains the cause of dead angle varying with points_per_second. (number of points get over before we complete 360 degree angle)

It should be something like:

delta_angle = 360/PointsToScanWithOneLaser

  for (auto Channel = 0u; Channel < ChannelCount; ++Channel)
  {
    for (auto i = 0u; i < PointsToScanWithOneLaser; ++i)
    {
      FVector Point;
      const float Angle = CurrentHorizontalAngle + delta_angle;
      if (ShootLaser(Channel, Angle, Point))
      {
        LidarMeasurement.WritePoint(Channel, Point);
      }
    }
}

I kindly request to fix this issue as many of us (autonomous vehicle researchers) are using carla because of it's support of lidar sensor.

[YashBansod]

@anshulpaigwar are you running your simulation in benchmark mode?
You should run the simulation in multiples of your Lidar Frequency to get a full 360 degree in one reading.
Alternatively, you could cache the Lidar reading to get a full 360 degree view.

The speed at which the server runs in non benchmark mode can vary a lot on a variety of factors.

[anshulpaigwar]

@nsubiron could you help with this issue of lidar sensor I still haven't been able to find a solution?

[nsubiron]

Hi @anshulpaigwar, as @YashBansod proposed you should run the simulator at fixed time-step ("benchmark" mode: ./CarlaUE4.sh -benchmark -fps=X). Check out #835 for more info.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.