slam-eval

A lightweight CLI tool and Python library for evaluating SLAM trajectory accuracy against ground truth.

Computes APE (Absolute Pose Error) and RPE (Relative Pose Error) from TUM-format trajectory files, with clean publication-ready plots.

---

Installation

pip install slam-eval

Or from source:

git clone https://github.com/your-username/slam-eval
cd slam-eval
pip install -e .

Requirements: Python ≥ 3.8, numpy, matplotlib, scipy — no ROS required.

---

Quickstart

CLI

# Basic evaluation (SE3 alignment by default)
slam-eval estimated.tum ground_truth.tum

# Save a 4-panel figure
slam-eval estimated.tum ground_truth.tum --plot results.png

# Sim3 alignment (for monocular visual SLAM)
slam-eval estimated.tum ground_truth.tum --align sim3 --delta 5

Output:

Loaded 'FAST-LIO2':     1240 poses, 62.0s, 48.32m
Loaded 'Ground Truth':  3720 poses, 62.0s, 48.18m

slam-eval results
  Estimated  : FAST-LIO2
  Ground truth: Ground Truth
  Poses      : 1240
  Alignment  : se3

APE (translation):
  RMSE:   0.0312 m
  Mean:   0.0271 m
  Median: 0.0244 m
  Std:    0.0163 m
  Min:    0.0011 m
  Max:    0.0981 m

RPE (translation, delta=1):
  RMSE:   0.0088 m
  ...

Python library

from slam_eval import load_tum, compute_rpe, AlignmentMethod, plot_trajectories

estimated    = load_tum("fast_lio2_run01.tum", name="FAST-LIO2")
ground_truth = load_tum("mocap_run01.tum",     name="MoCap GT")

result = compute_rpe(estimated, ground_truth, alignment=AlignmentMethod.SE3)
print(f"APE RMSE: {result.ape_translation.rmse:.4f} m")
print(f"RPE RMSE: {result.rpe_translation.rmse:.4f} m")

plot_trajectories(result, output_path="eval.png")

Comparing multiple algorithms:

from slam_eval import load_tum, compute_rpe

gt = load_tum("ground_truth.tum", name="MoCap")
algorithms = {"FAST-LIO2": "fast_lio2.tum", "DLIO": "dlio.tum", "RKO-LIO": "rko_lio.tum"}

print(f"{'Algorithm':<14} {'APE RMSE (m)':>14} {'RPE RMSE (m)':>14}")
print("-" * 44)
for name, path in algorithms.items():
    result = compute_rpe(load_tum(path, name=name), gt)
    print(f"{name:<14} {result.ape_translation.rmse:>14.4f} {result.rpe_translation.rmse:>14.4f}")

---

TUM format

timestamp  tx  ty  tz  qx  qy  qz  qw

Lines starting with # are comments. Compatible with evo and natively output by most ROS2 SLAM packages.

---

CLI reference

Option	Default	Description
--align	se3	none | se3 | sim3 | yaw_only
--delta	1	RPE step size in poses
--max-diff	0.02	Max timestamp diff for pose association (s)
--plot PATH	—	Save 4-panel evaluation figure
--heatmap PATH	—	Save standalone APE heatmap

---

Relation to evo

evo is the standard tool for SLAM evaluation and slam-eval is not a replacement. slam-eval is for cases where you want a single pip-installable dependency with a clean Python API — no ROS, no system dependencies, easy to embed in your own evaluation pipeline.

---

License

MIT