search

stereolabs / zed-ros-wrapper

ROS wrapper for the ZED SDK
https://www.stereolabs.com/docs/ros/
MIT License
447 stars 391 forks source link

Drift in ZED-M #449

Closed shakeebbb closed 4 years ago

shakeebbb commented 5 years ago

I have connected my ZED mini camera with Jetson TX2 and walked in a loop multiple times. I am visualizing the data in RVIZ. I cannot get rid of the drift. It seems like it's also not performing loop closures. The spatial memory is 'true' in the launch file and I've tried increasing and decreasing the frame rate and the resolution as well. Am I missing something especially for the loop closure part?

Screenshot from 2019-07-17 16-26-23

Myzhar commented 5 years ago

Hi @shakeebbb can I ask you about what version of the ZED SDK are you using?

shakeebbb commented 5 years ago

I'm using the latest setup available i.e., ZED SDK v 2.8.2 with Jetpack 4.2.

Myzhar commented 5 years ago

It seems a problem not strictly related to ROS, but more generally related to the SDK.

Can I ask you to write an email to support@stereolabs.com so we can open a ticket about it? If you can please record an SVO in the same conditions of the test you made and attach it to the email (better if you can provide a link to download it) and the conf file of your camera that you can find in the folder /usr/local/zed/settings/.

Thank you Walter

clungzta commented 4 years ago

Was there any resolution for this?

I seem to be experiencing the same issue. It appears that no loop closures are being detected (despite spatial_memory=true). I am experiencing significant drift. Thus, the mapping is near unusable.

It results in poor quality maps from ZED ROS Wrapper (map=true), and RTAB-Map.

I am using TX2 with ZEDM, 2.8.3 ZED SDK, Jetpack 4.2.1.

See images: Screenshot from 2019-09-16 20-59-42 Screenshot from 2019-09-23 20-01-58 Screenshot from 2019-09-24 20-18-14

(Note: I have the ZED head-mounted, and am seated rotating slowly and smoothly in my office chair to create those maps.)

obraun-sl commented 4 years ago

You can try with TrackingParameters::enable_imu_fusion at false to check if the loop closing is performing better.

shakeebbb commented 4 years ago

@clungzta I needed a quick solution for a lab project, I postponed the idea of using ZED odometry till the issue is resolved. Please let me know if ZED loop closure works out for you.

clungzta commented 4 years ago

Thanks. However, setting enable_imu_fusion=false had no discernible improvement. Following the earlier suggestion, I'll record an SVO and my conf file, and raise a support ticket.

Myzhar commented 4 years ago

Positional tracking has been improved with latest SDK versions. I close the issue