Trajectory Metadata & Config Parsing

[contagon]

This is quite a large PR that started as an effort to tackle the ugly large dictionary we were passing around with Trajectories by adding some types to it. It has expanded to include

• improved handling of searching for pipelines and datasets
• new config parsing that handles any errors a lot more gracefully (and Rust-like)
• new metadata classes to tack onto Trajectories. Should make handling of stats results a lot easier
• new stats commands to compute varying window sizes.
• better error handling when rerunning experiments

I'm going to sit on this PR a bit as I use it for research to make sure it's gets some run before getting merged

[contagon]

Also added in:

• revamp of stats cli module to use polars for filtering. This removed a lot of brittle code
• revamp of stats cli to be able to compute multiple rte windows in a single command
• Some misc doc improvements

[contagon]

• Sped up stats command significantly by moving hot code to C++ and avoiding using python's copy/deepcopy

[Copilot]

Pull Request Overview

This PR introduces a robust type system for trajectory metadata and improves configuration parsing throughout evalio. The key changes transform the architecture from using untyped dictionaries to strongly-typed metadata classes, along with introducing comprehensive error handling similar to Rust's Result pattern.

Key changes:

• Introduced typed Metadata and Experiment classes to replace untyped dictionaries for trajectory metadata
• Implemented parser modules for datasets and pipelines with comprehensive error handling
• Added support for window-based RTE computation in meters and seconds (previously fixed timesteps)

Reviewed Changes

Copilot reviewed 42 out of 43 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/test_stats.py	Removed metadata dictionary parameters from Trajectory initialization
tests/test_parsing.py	New test file for dataset and pipeline configuration parsing
tests/test_io.py	New test file for trajectory serialization/deserialization with metadata
tests/test_dataset_impl.py	Updated imports to use new dataset parser module
tests/test_csv_loading.py	Updated type annotations to use new GroundTruth metadata class
tests/test_cli_ls.py	Removed deprecated dataset building test
tests/make_test_data.py	Updated to use new dataset registration API
tests/get_lens.py	Updated to use new dataset registration API
python/evalio/utils.py	Added CustomException base class and pascal_to_snake utility function
python/evalio/types/extended.py	New file with Experiment and ExperimentStatus classes
python/evalio/types/base.py	New file with base Trajectory, Metadata, and GroundTruth classes
python/evalio/types/init.py	Reorganized type exports with new metadata classes
python/evalio/types.py	Removed (replaced by modular types/ directory)
python/evalio/stats.py	Updated for new window-based RTE computation and typed trajectories
python/evalio/rerun.py	Updated for new metadata types and visualization API
python/evalio/pipelines/parser.py	New pipeline configuration parser with error types
python/evalio/pipelines/init.py	Added parser exports
python/evalio/datasets/parser.py	New dataset configuration parser with error types
python/evalio/datasets/newer_college_2020.py	Removed type ignore comment
python/evalio/datasets/multi_campus.py	Fixed CSV field name and removed type ignore comment
python/evalio/datasets/base.py	Updated to use new metadata system and moved utility functions
python/evalio/datasets/init.py	Added parser exports
python/evalio/cli/writer.py	Removed (functionality moved to Trajectory.to_file)
python/evalio/cli/stats.py	Complete rewrite supporting new metadata and multiple windows
python/evalio/cli/run.py	Updated to use new experiment system with metadata
python/evalio/cli/parser.py	Removed (replaced by modular parser modules)
python/evalio/cli/ls.py	Updated to use new parser modules
python/evalio/cli/dataset_manager.py	Updated to use new dataset parser
python/evalio/cli/completions.py	Updated to use new parser modules
python/evalio/cli/init.py	Added custom module registration callback
python/evalio/init.py	Added automatic custom module registration from environment
pyproject.toml	Added polars and joblib dependencies
mkdocs.yml	Updated documentation settings
docs/ref/pipelines.md	Updated to explicitly list pipeline members
docs/ref/datasets.md	Simplified to show all dataset members
docs/quickstart.md	Updated example code to use new import patterns
cpp/evalio/types.h	Added SE3::error and SE3::distance helper methods
cpp/bindings/types.h	Added Python bindings for new SE3 methods and copy constructors
cpp/bindings/ros_pc2.h	Added parse_csv_line and closest helper functions
cpp/bindings/main.cpp	Disabled nanobind leak warnings
README.md	Updated with citation information and simplified quick start
CITATION.bib	New citation file for arXiv paper

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Using lambda with noqa to bypass linting is acceptable here, but the filter_method on line 293 uses eval() which is a security risk. The eval() call allows arbitrary code execution through the filter_str parameter. Consider using a safer alternative like ast.literal_eval() or a custom expression parser that doesn't allow arbitrary code execution.