How to use evaluation on a gazebo simulation

[mikexyl]

Hi @Schmluk , after successfully tested mav_active_3dplanning with our work, I also want to use the evaluation app. But I am a liitle confused about how to connect that evaluation_data_node to gazebo world topics. And is this the case, if I do not run that evaluation node, the planner node itself won't generate enough evaluation data?

[mikexyl]

And program crashed when I try to set a bigger min_path_length, because I am running it in a outdoor simulation, and 0.5 looks too small compared to environment size and mav speed.

This error showed up both with min path length 5 and 10. How should I set params to make each path branch longer?

[Schmluk]

Hi @LXYYY,

Good to hear. Regarding the first question, the planner node itself only stores data for the performance analysis (e.g. where how much time was spent). The eval_data node manages simulation setup and shutdown and periodically stores the map for later evaluation against ground truth pointclouds. To run it with Gazebo, the following topics need to be remapped:

• ue_out_in: Counts how many pointclouds were received. You can also leave this blank if it does not interest you.
• collision: When this topic receives a message (of type std_msgs/String) the experiment is shutdown and collision is entered to the data log so you can check what happened in your experiments. I don't know whether this information is readily available in your simulation, if the simulation cannot collide or detect collision you can also leave this empty and it will always run till maximum time is reached.

Regarding your second question, I think max_path_length is the param you want to increase, in any case make sure it is larger than min_path_length. If the error persists can you please add the full configuration to the issue?

Thank you and best wishes, Lukas

[mikexyl]

Thanks @Schmluk , a bigger max_path_length works for me. I will try the evaluation later. But I run into another error today, when I was trying to set trajectory generator type to FeasibleRRTStar to get smoother derivatives, it crashed: and I print trajectory.time_from_start_ns before crash, it showed and I tried to start planning when last trajectory generated by another node, both finished or unfinished, the error still exists. Any idea?