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

Autoware: Current Status of HD map localization?

asked 2020-06-30 20:42:01 -0500

doodleshr gravatar image

It seems HD map is not used for localizaiton in, and I’m wondering if HD map is used for localization in the upcoming

Hi guys,

Sorry if I’m posting at the wrong place.

There are some HD map related questions that I hople someone can shed some light on:

  1. can localization be done with only HD map (semantic informations such as lanes, traffic signs etc)?
  2. what map format is used for the current stack implementation? Lanelet? Apollo OpenDrive? Standard OpenDrive?
  3. is there a wiki/tutorial/ about HD map localization?

Thanks for trying to help Joey

edit retag flag offensive close merge delete

1 Answer

Sort by » oldest newest most voted

answered 2020-07-06 15:11:16 -0500

Josh Whitley gravatar image

@doodleshr You cannot use an "HD Map" (specifically the semantic portion that you mention) for localization because there are not enough physical landmarks in the environment to accurately reference your location in the semantic map. Let's say you're on a long, straight stretch of road heading North with the same speed limit for the entire section. If you localize only using the lane markings on your right and left (East and West), how would you know how far North or South you are?

As for the HD map format, we will be using Lanelet2. I say "will be" because this functionality is still being implemented in our current milestone in Autoware.AUTO. You can follow along with our progress in this issue.

edit flag offensive delete link more



Thanks for your explanation.

I agree with you on the special case that a vehicle is running on a long straight road, the vehicle cannot localize itself under this situation.

But for some more common cases such as in the city, where traffic signs, poles and road lanes are every where, and most importantly GNSS is not likely to work because of all the buildings and bridges. I think it might be a good idea to utilize these semantic informations.

I recently found some works, which showcased some error analysis of semantic map localization and here I quote two: 1. AVP-SLAM: Semantic Visual Mapping and Localization for Autonomous Vehicles in the Parking Lot; 2. Exploiting Sparse Semantic HD Maps for Self-Driving Vehicle Localization

I am still trying to make a demo based on the idea, so I am very likely to be over positve. I'd love to hear more ...(more)

doodleshr gravatar image doodleshr  ( 2020-07-07 21:12:16 -0500 )edit

@Joey, Indeed it was a nice idea to utilize the HD maps to localize the vehicle. Currently am also working in the same direction. Currently, Autoware localization heavily depends on the Point cloud map for precise (centimeter-level ) localization, but generating and updating the Point cloud maps is quite expensive and time-consuming. Also, failure detection with the LiDAR matching algorithm is difficult. A combination of HD map and Point cloud will greatly benefit the localization. If you are interested, we can discuss and share more about this. Regards, Ajay

Ajay gravatar image Ajay  ( 2020-09-22 03:45:39 -0500 )edit

Hi guys do you know how to localize the vehicle in a huge PCD map, such as SanFranscisco: If you have some idea please give me a suggestion here:

Vini71 gravatar image Vini71  ( 2021-08-09 13:32:27 -0500 )edit

Question Tools

1 follower


Asked: 2020-06-30 20:42:01 -0500

Seen: 963 times

Last updated: Jul 06 '20