difference between conventional ndt and this package

[weisongwen]

hi @koide3

thanks for your kind effort on this package, could you tell me what is the difference between this package and conventional ndt ?

Best Welson

[koide3]

Hi, The major difference is that this package provides a capability of UKF-based sensor pose estimation. It allows us to start NDT with a good initial guess and makes the scan matching fast and robust.

[weisongwen]

Hi, @koide3 Thanks for your kind reply have you try to integrate the GPS into the localization?

Best Welson

[koide3]

Hi, No, at this time, we've not integrated GPS info with localization. We use GPS at only the mapping step.

[weisongwen]

Hi, @koide3 thanks for your kind reply, in the mapping step. can this package can do real time mapping? Thanks Best Welson

[koide3]

Hi, Yes. We confirmed that this package can run real-time on a standard desktop PC. We expect that it can run on a laptop PC as well, but it depends on the setting.

[weisongwen]

Hi @koide3 Thanks for your kind reply, I mean: if the mapping package "hdl_graph_slam" can run real time? Thanks Best Welson

[koide3]

Hi, Yes, I meant "hdl_graph_slam" can run real-time.

[weisongwen]

hi @koide3
thanks for your kind reply, best welson

[jeannotes]

@weisongwen Hi, did you run the bag? I didn't not get the final result as in the viseos.

[weisongwen]

Hi， @JSRGFJZ I have not test the localization package yet, Best Welson

[koide3]

Hi, @JSRGFJZ How did it fail? Did the programs start correctly?

[jeannotes]

@koide3 Hi, I notice that when I didn't using IMU to predict, the error is very large.WHY? is it because of only the lidar is one line instead of multi-line? if I have velodyne-16, could it be better?

[weisongwen]

Hi @JSRGFJZ
I think that you may not give a good initialization when you begin the mapping based localization process.

[koide3]

Hi @JSRGFJZ , You are trying to run the bag (hdl_400.bag, 32 line LIDAR), right? I guess the computation power of your PC is insufficient for the default localization parameters. Can you change ndt_neighbor_search_method in hdl_localization.launch to "DIRECT1" and re-try to run the bag? It makes the localization faster significantly.

[jeannotes]

@koide3 Actually I already set it to DIRECT1, but the result is not good.

[koide3]

@JSRGFJZ Sorry for the late response. I still can't understand how it fails. Can you try it with a slower playback speed? Like:

rosbag play --clock -r 0.1 hdl_400.bag

[jeannotes]

@koide3 I already tried that, it still doesn't work. I didn't use IMU, this should be the reason. but I read your thesis, you said it also could, so should I alter some parameters?

[koide3]

@JSRGFJZ Yes, it should be able to use it without IMU. But, the lack of IMU makes the localization weak against quick motions. Make me sure that it works with IMU, right?

[jeannotes]

Yes, it works with IMU. I don't know if it is the only situation with me, but I really appreciate your work! Thank you.

j.z.feng@foxmail.com

From: koide3 Date: 2018-04-15 22:04 To: koide3/hdl_localization CC: Jean; Mention Subject: Re: [koide3/hdl_localization] difference between conventional ndt and this package (#1) @JSRGFJZ Yes, it should be able to use it without IMU. But, the lack of IMU makes the localization weak against quick motions. Make me sure that it works with IMU, right? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.