Add RPE computation for different segment lengths, add basic evaluation example

[mitchellcohen3]

No description provided.

[Copilot]

Pull request overview

Adds trajectory evaluation utilities (ATE + RPE over multiple segment lengths) and a runnable example demonstrating how to compare multiple SLAM/odometry estimates on a dataset.

Changes:

• Introduces navlie.utils.evaluation with ATE computation, RPE computation over configurable segment lengths, and an RPE boxplot helper.
• Adds examples/ex_evaluation.py to demonstrate loading trajectories, simulating drift, computing metrics, and plotting results.
• Updates README Python version badge.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 6 comments.

File	Description
navlie/utils/evaluation.py	New evaluation + plotting helpers built on evo and seaborn/pandas.
examples/ex_evaluation.py	New end-to-end example computing ATE/RPE and plotting trajectories + RPE distributions.
README.rst	Updates the Python version badge range.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

plot_rpe_boxplot assumes the dataframe is non-empty (e.g., df["segment_length"]), but compute_rpe_over_segment_lengths can return empty arrays on failure, which can lead to rows=[] and an empty dataframe. In that case this will raise a KeyError. Consider handling the empty/no-data case explicitly (e.g., early return with an empty plot or a clear error).

Suggested change

	df = pd.DataFrame(rows)
	if ax is None:
	fig, ax = plt.subplots(figsize=figsize)
	else:
	fig = ax.get_figure()
	df = pd.DataFrame(rows, columns=["Estimator", "segment_length", "error"])
	if ax is None:
	fig, ax = plt.subplots(figsize=figsize)
	else:
	fig = ax.get_figure()
	if df.empty:
	ax.set_xlabel(xlabel)
	ax.set_ylabel(ylabel)
	fig.tight_layout()
	return fig, ax

[Copilot]

Inside the estimator loop, gt_states is reassigned from associate_and_align_trajectories(...). This means the second iteration no longer uses the original full ground-truth trajectory, which can change both plotting and metrics depending on association results of the first estimate. Use a separate variable (e.g., gt_states_aligned/gt_states_sync) so each estimate is associated against the same original groundtruth.

[Copilot]

This example aligns trajectories via associate_and_align_trajectories(...) but then computes ATE/RPE using est_states (unaligned) while also passing align=True into compute_ate / compute_rpe_over_segment_lengths. This duplicates work and makes it unclear which alignment the reported metrics correspond to. Consider either (a) computing metrics on est_states_aligned with align=False, or (b) removing the explicit alignment step and relying on the metric functions for alignment (keeping a separate aligned copy only for plotting if needed).

[Copilot]

New evaluation utilities are introduced here without accompanying unit tests. Since the repo already has a pytest suite, consider adding basic tests (e.g., identical trajectories -> ~0 ATE/RPE; synthetic drift -> RPE arrays non-empty) to validate the evo integration and prevent regressions.

[Copilot]

sns.boxplot is being called with fill and gap defaults, but the package dependency is seaborn>=0.11.2 (setup.py). These kwargs are not supported in older seaborn versions and will raise a TypeError for users pinned near the minimum. Either remove/avoid these kwargs (use only options available in 0.11.x), or bump the seaborn minimum version accordingly so the API matches.

Suggested change

	# Optional boxplot kwargs
	kwargs = dict(
	fill=False,
	width=0.8,
	gap=0.15,
	# Optional boxplot kwargs. Keep defaults compatible with seaborn>=0.11.2.
	kwargs = dict(
	width=0.8,

[Copilot]

The docstring lists a title: str parameter, but the function signature doesn't accept title. This is misleading for users; either add the parameter and apply it (e.g., ax.set_title(...) / fig.suptitle(...)) or remove the docstring entry.

Suggested change

	title: str
	Plot title.