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
862 stars 139 forks source link

Load and Localize #98

Open tom-deBach opened 1 year ago

tom-deBach commented 1 year ago

Hi, I have some questions about the LoadMapAndLocalize action. I tried it several times and found it quite hard to correctly pinpoint the location with the prior guess of the pose, so the map would not load (The best result I got was circa 1.5m radius around the correct position).

Am I doing something wrong or do I really have to know my position this close to load a map? I would have expected, that it maybe takes more time if the guess is not too good, or takes more landmarks than the current into account.

I found this parameter, but it seems, this is still in development: Parameter_cuVSLAM_status radius_parameter

This seems to include, what i would hope for. Are there current plans to implement this?

Kind regards and thanks in advance

jaiveersinghNV commented 1 year ago

We're working on this feature, but we don't have any updates on when it will be delivered. Could you try using a feature in your task environment to align the start of your mapping and then re-align the robot there during the load map stage?

tom-deBach commented 1 year ago

First of all: Thanks for your work :) I really like the packages and combination with VSLAM, Nvblox and nav2.

Unfortunately I don't have such a feature. Our goal is to load a map into a electronic travel aid for visually impaired people, so we cant guarantee a "starting position" as you would have with an AGV for example and as it would be needed to localize with the "load and localize" function, if I understood correctly.

Could the isaac_ros_map_localization package be of use with a stereo camera, instead of a lidar?

And a small heads up: This does not really concern you at nvidia directly as it is more of a Intel topic, but we had problems with the auto exposure not working together with the laser_emitter_on_off function of the realsense splitter on a Realsense D435.

Kind regards

jaiveersinghNV commented 1 year ago

The isaac_ros_map_localization package is currently lidar-only, and it will not work with a stereo camera. We're working on speeding up the LoadMapAndLocalize action to make it feasible for you to run a grid search over multiple candidate initial guesses.

If you're experiencing issues with the RealSense camera, you may be able to find better support for it on the Intel forums. Note that the RealSense splitter we provide is explicitly designed to support simultaneous usage of VSLAM and Nvblox; if you're only using one of the two, you may be able to avoid the problem by removing the splitter from your graph entirely.