ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

Autoware.Auto Localization gets lost when turning in certain spots of the map

asked 2021-07-12 14:14:01 -0600

RDaneelOlivaw gravatar image

Hi,

I'm having this issue that when I'm using the localization system in Autoware.Auto, the car gets lost when it turns too fast, enters a relatively long corridor or just plain whenever it wants.

VIDEO Of localization failing at the end

I thought it was the PCL that wasn't high def enough but I generated a more detailed pcl with as far as I see enough details to make a good localization, but it just gest lost.

It seems its a reported issue here: ISSUE But it's a YEAR OLD post so I thought that would have improved.

So I'm using these files:

localizer.launch.py

ndt_localizer.param.yaml

map_provider.launch.py

map_publisher.param.yaml

My question is then:

1) How could I improve this so that the localization works ok. Any suggestions on changing sampling rate, voxel sizes, using an alternative to Autoware.Auto localization system?

2)Why aren't we using GPS and other sensors for localization improvement? the p2d_ndt_localizer_exe I don't see anywhere its using any data except the PCL and the lidar sensor.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2021-07-17 13:31:45 -0600

Josh Whitley gravatar image

Two answers:

  1. Running the simulator on the same machine as Autoware.Auto is not recommended for this reason. NDT is a very compute-heavy algorithm and SVL is also very compute-heavy. This can cause significant latency in the output of NDT, causing the next pass to be much more likely to be unable to match.
  2. Just recently, we added a state estimator and gave it the ability to use GNSS data as input. However, we are still working on a good demonstration. The pipeline would basically be that the state estimator takes in the pose estimate from NDT and also the position from GNSS which is converted from a sensor_msgs/msg/NavSatFix to an autoware_auto_msgs/msg/RelativePositionWithCovarianceStamped using the gnss_conversion_node.
edit flag offensive delete link more

Comments

One additional question regarding your first answer:

  • Assuming you have a powerful machine, say 32-core Ryzen with 64 GB RAM - would this recommendation of having a distributed computing setup still hold? In other words, can I assume that I'm fine if I'm running the simulation on the powerful machine?
dmajstorovic gravatar image dmajstorovic  ( 2021-11-14 10:53:28 -0600 )edit

Question Tools

2 followers

Stats

Asked: 2021-07-12 14:14:01 -0600

Seen: 286 times

Last updated: Jul 17 '21