search

NVIDIA-ISAAC-ROS / isaac_ros_visual_slam

Visual SLAM/odometry package based on NVIDIA-accelerated cuVSLAM
https://developer.nvidia.com/isaac-ros-gems
Apache License 2.0
816 stars 126 forks source link

How to improve performance on Jetson Xavier NX? #95

Closed fxprime closed 11 months ago

fxprime commented 1 year ago

Hi. Thank you for great visual slam.

I begin with confuse about frame orientation. But now i can manage to run it properly(atleast i guess) The problem is, while it run with rectified images from machine transport to docker that isaac workspace is located, it seem not smooth as expect. So i record a bag file below then move to docker folder. But it improve just a little bit. The problem still occur with sometime have smooth trajectory. Sometime struggle along the way. as show in this video

https://youtu.be/ssxxNdnOZs8

Bagfile https://shorturl.at/eGZ36

And this is launch file with configuration

https://gist.github.com/fxprime/94c28b81890c24c8642b841b8b9136f9 https://gist.github.com/fxprime/bbae3381174efd4cdab870364eca852f

I don't sure it's machine limitation or some cause. Here is machine information

fx@ubuntu:~/jetsonUtilities$ python3 jetsonInfo.py 
NVIDIA NVIDIA Jetson Xavier NX Developer Kit
 L4T 35.3.1 [ JetPack UNKNOWN ]
   Ubuntu 20.04.6 LTS
   Kernel Version: 5.10.104-tegra
 CUDA 11.4.315
   CUDA Architecture: 7.2
 OpenCV version: 4.5.5
   OpenCV Cuda: YES
 CUDNN: 8.6.0.166
 TensorRT: 8.5.2.2
 Vision Works: NOT_INSTALLED
 VPI: 2.2.7
 Vulcan: 1.3.204

Can someone help me to improve the performance or something i missed? Thank you advanced.

ripdk12 commented 1 year ago

This is a long shot, but are you running this from an SD card? I personally experienced quite a performance boost after switching to SSD.

I also use a Xavier NX and I get much better performance, often 90hz VIO data.

tdnet12434 commented 1 year ago

This is a long shot, but are you running this from an SD card? I personally experienced quite a performance boost after switching to SSD.

I also use a Xavier NX and I get much better performance, often 90hz VIO data.

OMG. 90hz is much more than I expected. Are you also running on docker? If so, it's impressive! I also use SSD run overall Ubuntu (installed directly from sdkmanager to SSD) Still can't figure out why performance is worse like this.

jaiveersinghNV commented 1 year ago

Based on one of your launchfiles, it seems like you might be running two instances of Visual SLAM concurrently? If so, have you tried running just one instance and observing the performance there?

You might also install the jetson_stats tool to monitor the system utilization during the operation of your launch graph. It could be the case that some non-Isaac ROS background process is consuming too much CPU or memory bandwidth.