Skip to main content

Module slam_benchmark

Module slam_benchmark 

Source
Expand description

Relative-pose SLAM metric of Kümmerle et al., “On measuring the accuracy of SLAM algorithms” (Autonomous Robots, 2009).

Instead of comparing a trajectory against a global ground truth, the benchmark provides relations: verified relative poses δ*ᵢⱼ between the poses at two timestamps. For an estimate the error of each relation is δᵢⱼ ⊖ δ*ᵢⱼ with δᵢⱼ = xᵢ⁻¹ ⊕ xⱼ, and translational and rotational parts are averaged separately (absolute and squared). The metric is invariant to the global frame, so any SLAM output can be scored.

Relations files have one relation per line:

timestamp_i timestamp_j x y z roll pitch yaw

Structs§

Relation
A reference relative pose between the poses at two timestamps.
RelationErrors
Aggregated relation errors.

Functions§

evaluate_relations
Scores trajectory (timestamped poses, any order) against relations.
format_relations
Formats relations in the benchmark file format.
parse_relations
Parses a relations file; blank lines and # comments are skipped.