Skip to content

Quickstart

This tutorial provides minimal runnable examples for the three modules to verify your setup. For detailed usage, see each module's user guide.

Path placeholders

Replace /path/to/scenes/ or /path/to/raw_scenes/ with your actual data directory. For centralized asset management, use $NAVARENA_DATA_DIR/assets/{dataset}/{scene_id}/ (e.g. $NAVARENA_DATA_DIR/assets/x2robot/17dc3367/).

End-to-End Workflow

The three modules execute sequentially; each step's prerequisites are the outputs of the previous step: raw PLY → V1 assets → Episodes → eval results.

Asset Preprocessing

Prerequisites: Raw 3DGS PLY files (e.g. {scene_dir}/scene.ply).

Command:

cd navarena-forge
python -m navarena_forge run-pipeline \
    --config navarena_forge/configs/pipeline.yaml \
    --scene-dir /path/to/scenes/17dc3367 \
    --source-dataset x2robot

Expected output: The pipeline writes in-place to --scene-dir: aligned.ply, nav_map.pgm, manifest.json, nav_map.yaml, nav_mask.png, etc.

Asset centralization

Output does not automatically go to $NAVARENA_DATA_DIR/assets/. To centralize assets, place scene directories under $NAVARENA_DATA_DIR/assets/ and pass the corresponding paths. For batch processing, see Asset Preprocessing CLI.

Data Generation

Prerequisites: Completed preprocessing (V1 assets), scenes under $NAVARENA_DATA_DIR/assets/.

Command:

cd navarena-gen
python scripts/generate_data.py --config configs/examples/pointnav_example.yaml

Expected output: Episodes and trajectories written to $NAVARENA_DATA_DIR/datasets/ (or config-specified output path). Check for meta/episodes.parquet and data/chunk-*/trajectories.parquet.

Use --parallel --num-workers 4 --batch-size 20 for parallel generation. See Data Generator Configuration.

Evaluation

Prerequisites: V1 assets + Episodes conforming to evaluation data format.

Command:

cd navarena-bench
python -m navarena_bench.scripts.eval --config configs/eval/default_eval.yaml

Expected output: Evaluation results in eval_results/ (or config-specified path). Metrics (SR, SPL, etc.) printed to console.

Replay: python scripts/replay_eval.py --results eval_results/ --output replay.mp4. See Evaluators.

Next Steps