Skip to content

Latest commit

 

History

History
14 lines (13 loc) · 585 Bytes

dynamicEnv_talk.md

File metadata and controls

14 lines (13 loc) · 585 Bytes

Dynamic Environments in Least Squares SLAM

Cyril Stachniss

link : https://www.youtube.com/watch?v=vtnh9TExdOI&t=2833s


Notes:

  • Gaussian distribution doesn't take outliers into account
  • Huber loss is better, downweights large errors
  • Adaptive loss function is even better, but depend on a alpha parameter
  • Optimisation pipeline that alternates iterations between alpha and whole state
  • impressive results on ICP with moving objects
  • Semantic maps on LiDAR data with the range image in a CNN
  • SLAM with semantic ICP exploits semantic to remove or ignore moving objects