search

MIT-SPARK / Kimera-Multi-Data

A large-scale multi-robot dataset for multi-robot SLAM
MIT License
139 stars 10 forks source link

Question about start times and RPY of ground-truth #11

Open TakShimoda opened 9 months ago

TakShimoda commented 9 months ago

Hello, thank you for this dataset.

I've been trying to test this dataset on multi-robot SLAM systems outside of Kimera-Multi, and I had some questions about the dataset:

  1. After performing trajectory alignment with the ground-truth, should that result have its Roll angle adjusted by 90 degrees? From another issue, one of the author does confirm it's about 90 degrees, but I noticed it depends on the robot, and sometimes I also have to subtract 270 degrees from the yaw to get accurate results. Adjusting this significantly affects the RPE performance.
  2. Is it okay to fast forward times in the bag files? I noticed, especially for 1208 campus hybrid, the robots stay still for a long time at the beginning and end and I was wondering if these were necessary? For example, I'm trying to use ORB-SLAM3 as one of the front-ends in a multi-robot setup and it's recommended to have IMU excitation in the beginning frames.
  3. Considering this is a very large dataset, do you think there would be communication bandwidths issues for centralized systems? I noticed experiments with individual robots can get good results, but with all 8 (on 1208 for example) some of the trajectories aren't so good, so I was thinking of running sequentially.

Thank you