ros controller crashes with ToF/3D cam simulation

[machinekoder]

Describe the Bug I would like to simulate a 3D/ToF camera. For this purpose I've set up a sim with with a Lidar device. The ToF cam has a resolution of 224x171 and FOV of 62x45deg, I've entered that in Webots. So far so good.

When I now try to use the sim with a ros controller, the controller just crashes. I suspect that this is due to the high number of layers since it doesn't crash when I reduce the layers.

Steps to Reproduce

1. Create a robot with a Lidar device
2. Set the Lidar device horizontalCameraResolution to 224 and numberOfLayers to 171.
3. Set the robots controller to ros
4. Start the sim and enable both the lida and enable_point_cloud
5. The ros controller crashes.

Expected behavior The ros controller works as expected or gives an appropriate error message.

Screenshots none

System

• Operating System: Ubuntu Bionic 18.04
• Graphics Card:

  
  glxinfo -B
  name of display: :0
  display: :0  screen: 0
  direct rendering: Yes
  Memory info (GL_NVX_gpu_memory_info):
  Dedicated video memory: 4096 MB
  Total available memory: 4096 MB
  Currently available dedicated video memory: 2878 MB
  OpenGL vendor string: NVIDIA Corporation
  OpenGL renderer string: Quadro P2000/PCIe/SSE2
  OpenGL core profile version string: 4.6.0 NVIDIA 410.104
  OpenGL core profile shading language version string: 4.60 NVIDIA
  OpenGL core profile context flags: (none)
  OpenGL core profile profile mask: core profile

OpenGL version string: 4.6.0 NVIDIA 410.104 OpenGL shading language version string: 4.60 NVIDIA OpenGL context flags: (none) OpenGL profile mask: (none)

OpenGL ES profile version string: OpenGL ES 3.2 NVIDIA 410.104 OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20

**Additional context**
Webots sim: https://wolke.roessler.systems/s/q3Map9kMnPxwYA6 `simple.wbt`

[DavidMansolino]

I can indeed reproduce the issue, it crashes with the following stack:

Thread 1 "ros" received signal SIGSEGV, Segmentation fault.
0x000000000049d733 in RosLidar::publishPointCloud (this=0x1d46e70)
    at RosLidar.cpp:132
132     point.z = pointCloud[i].z;
(gdb) where
#0  0x000000000049d733 in RosLidar::publishPointCloud (this=0x1d46e70)
    at RosLidar.cpp:132
#1  0x000000000049d0cf in RosLidar::publishAuxiliaryValue (this=0x1d46e70)
    at RosLidar.cpp:97
#2  0x00000000004c883b in RosSensor::publishValues (this=0x1d46e70, step=0)
    at RosSensor.cpp:84
#3  0x00000000005a8257 in Ros::run (this=0x1d2e730, argc=1, 
    argv=0x7ffc48afa5b8) at Ros.cpp:454
#4  0x00000000005d90c5 in main (argc=1, argv=0x7ffc48afa5b8) at main.cpp:19

Here are the ros command to reproduce:

rosservice call /robot_5563_cyberbotics_david/lidar/enable "{value: 32}"
rosservice call /robot_5563_cyberbotics_david/lidar/enable_point_cloud "{value: true}"
# already crashed
rostopic echo /robot_5563_cyberbotics_david/lidar/laser_scan/layer0

[DavidMansolino]

We have been able to isolate the problem in a simpler controller not related with ROS:

#include <webots/robot.h>
#include <webots/lidar.h>

#include <stdio.h>

#define TIME_STEP 64

int main(int argc, char **argv) {
  wb_robot_init();

  WbDeviceTag lidar = wb_robot_get_device("lidar");
  wb_lidar_enable(lidar, TIME_STEP);
  wb_lidar_enable_point_cloud(lidar);

  while (wb_robot_step(TIME_STEP) != -1) {
     int i = 0;
     const WbLidarPoint *point_cloud = wb_lidar_get_point_cloud(lidar);
     for (i = 0; i < wb_lidar_get_number_of_points(lidar); ++i) 
       printf("%d %lf\n", i, point_cloud[i].x);
  };

  wb_robot_cleanup();

  return 0;
}

With this simple world file:

#VRML_SIM R2019b utf8
WorldInfo {
}
Viewpoint {
  orientation -0.6931032800836722 0.6931032800836722 0.1980295085953349 0.75
  position 1.2 1.6 2.3
}
TexturedBackground {
}
TexturedBackgroundLight {
}
RectangleArena {
}
Robot {
  children [
    Lidar {
      translation 0 0.12 0
      horizontalResolution 224
      verticalFieldOfView 1.6
      numberOfLayers 171
    }
  ]
  controller "my_controller_lidar"
  controllerArgs "--name=robot"
}

[fabienrohrer]

I have the same crash on macOS.

[fabienrohrer]

This is due to a recent regression introduced in the migration to the POSIX shared memory #872 .

https://github.com/cyberbotics/webots/blob/revision/src/lib/Controller/api/abstract_camera.c#L54 does the wrong assumption that the shared memory is identical in every devices. For the lidar, the point cloud is concatenated.

Computing this size correctly should fix the issue.

@machinekoder Meanwhile we provide a good fix, you could certainly revert to R2019b.rev0 to solve this issue.

[DavidMansolino]

well spotted @fabienrohrer !

[fabienrohrer]

@machinekoder This bug has been fixed in #1091. The patch will be included in the revision nightly build of tomorrow: https://github.com/cyberbotics/webots/releases

[fabienrohrer]

Any feedback is welcome.