New method helps robots find their way in changing environments

A method uses sparse anchors to register a complete trajectory to a reference map.

Repeated robotic monitoring requires localization across visits to the same environment. This is difficult if the environment changes.

Changes can reduce persistent visual content. Repetitive structures can also lead to incorrect cross-session correspondences.

The approach relies on a sparse set of reliable absolute pose constraints. It combines these with within-session relative-pose information.

Absolute alignment hypotheses are estimated through visual place recognition, local feature matching, and geometric verification. Sequence consensus rejects inconsistent hypotheses to retain absolute anchors.

These anchors are refined through featuremetric alignment against a Gaussian feature field. They are combined with relative-pose constraints in a robust pose-graph optimization.

In 15 experiments using the HILTI26 construction-site dataset, the method achieved 94.4% recall within 1m1\text{m} and 100.0% within 5m5\text{m}.

Key facts

  • The method achieves 94.4% recall within 1m1\text{m} and 100.0% within 5m5\text{m}.
  • It was evaluated on 15 cross-session experiments from the HILTI26 construction-site dataset.
  • The approach uses sparse, reliable absolute anchors to register a complete trajectory.

All sections

Edition