feat: add ROS-free standalone test, MuJoCo DER evaluation, and update .gitignore

[muhammad-mahad]

Summary

TrackDLO currently requires a physical robot setup (ROS, RealSense camera) to run any evaluation. This PR adds two hardware-free scripts and a .gitignore update.

standalone_test.py

A minimal ROS-free smoke test for the trackdlo_cpp pybind11 backend.

• Synthesises a catenary rope drooping across N frames
• Samples dense point clouds with configurable Gaussian noise
• Runs the full reg → sort_pts → cpd_lle pipeline and reports per-frame tracking error vs ground truth
• Dependencies: numpy, matplotlib, and the compiled trackdlo_cpp.so already in pybind_ext/ — nothing else required
• Result on default config: mean 16 mm, max 21 mm, ~41 ms/frame

python standalone_test.py
python standalone_test.py --frames 30 --noise 0.005 --visualize-3d

eval/mujoco_eval.py

A higher-fidelity evaluation using the adapteddlo_muj Discrete Elastic Rod (DER) simulator (MuJoCo 3.x).

• Builds a physics-accurate DER rope with real gravity (bending + torsion)
• Moves one endpoint sinusoidally over the sequence (0 → 20 cm → 0)
• Samples point clouds from DER segment surfaces
• Measures TrackDLO accuracy against DER ground truth
• Dependencies: mujoco ≥ 3.0, numpy, matplotlib, adapteddlo_muj
• Result on default config: mean 16 mm, max 37 mm (at peak lift), ~19 ms/frame on a 0.8 m / 30-segment cable

python eval/mujoco_eval.py
python eval/mujoco_eval.py --adapteddlo-path /path/to/adapteddlo_muj
python eval/mujoco_eval.py --frames 30 --r-len 1.0 --r-pieces 40

.gitignore

Added *.pdf and MUJOCO_LOG.TXT to suppress generated outputs from both scripts.

Test plan

• python standalone_test.py → runs to completion, prints mean/max error, saves trackdlo_standalone_results.pdf
• python eval/mujoco_eval.py → runs to completion, prints mean/max error, saves eval/mujoco_eval_results.pdf
• git status shows no new untracked files after both runs

[muhammad-mahad]

Following up with a bit more context in case it's useful. These two scripts are fully additive — standalone_test.py and eval/mujoco_eval.py — and they don't touch any existing tracking code or change the ROS pipeline, so the diff should be low-risk to merge.

The motivation: I wanted to be able to exercise the registration → sorting → CPD pipeline without a robot or a ROS stack, so it's reproducible on any machine. standalone_test.py runs synthetic catenary sequences with configurable noise; eval/mujoco_eval.py drives a physically realistic rope using adapteddlo_muj's Discrete Elastic Rod simulator (gravity + endpoint manipulation) and measures tracking error against the simulated ground truth. On a 0.8 m / 30-segment cable I get ~16 mm mean error at ~19 ms/frame.

Happy to split this into a smaller PR (just the standalone test) or adjust to fit the repo's conventions — whatever's easiest to review. @tbretl no rush, just flagging since the repo's been quiet; glad it was useful to reproduce TrackDLO either way.