Learning-Deep-Learning

Towards lifelong feature-based mapping in semi-static environments

December 2020

tl;dr: Feature persistence model to keep features in the map up to date.

Overall impression

Vanilla SLAM assumes a static world. They have to adapt in order to achieve persistent autonomy. This study proposed a feature persistent model that is based on survival analysis. It uses a recursive Bayesian estimator (persistence filter).

In summary, any observation existence boosts the existence confidence, any observation of absence degrades existence conf, and lack of observation decays existence conf.

This method has a good formulation but seems to be a bit heavy and does not allow large scale application. See Keep HD map updated.

Key ideas

Technical details

Notes