EuRoC Forum

[sven]

Dear EuRoC team, please correct me if I am wrong. Currently each simulation run for task 3 and 4 produces different sensor noise. Two executions of the same solution for this tasks will result in two different outcomes. Do you plan to run the submitted solutions multiple times and calculate a mean over the reached points? Otherwise this means, each team will have different conditions.

[markusachtelik]

dear sven,

very good question! In the simulation, there is a "random (i.e. system time)" seed for the random number generator, which produces a different outcome in every run. This behaviour is intended, since there are never going to be the same noise conditions in a real system and contestant solutions should be able to handle this.

For the final evaluation, we will choose a fixed seed such that starting conditions are the same for each team. If you want to fix the seed for debugging purposes, you can do the following:

Add this line

Code: Select all

<randomEngineSeed>YOUR_COSTOM_INTEGER_SEED</randomEngineSeed>

to euroc_simulation_server/urdf/firefly_tX_X.gazebo.xacro
between the

Code: Select all

<plugin filename="libmav_gazebo_pose_plugin.so" name="pose_sensorX"> ... </plugin>

tags for the pose_plugin(s), and

in euroc_simulation_server/urdf/firefly_base.urdf.xacro between the

Code: Select all

<plugin filename="libhector_gazebo_ros_imu.so" name="imu_ros"> ... </plugin>

tags