apairo-preprocess

Preprocessing pipelines for apairo datasets — LiDAR odometry and traversability ground truth generation.

---

Installation

pip install git+https://github.com/apairo-robotics/apairo_preprocess.git

Optional dependencies:

pip install kiss-icp   # for KissICPOdometry
pip install open3d     # for GICPOdometry

Requires Python ≥ 3.11.

---

Preprocessors

Odometry

Class	Output channel	Backend	Output
KissICPOdometry	kissicp_poses	KISS-ICP	(4, 4) float64 pose per scan
GICPOdometry	gicp_poses	Open3D GICP	(4, 4) float64 pose per scan

Traversability

Class	Output channel	Method
TraversabilityFromLabels	trav_label	Maps semantic class IDs to binary traversable/non-traversable
TraversabilityFromTrajectory	trav_gt	Labels points inside the robot's forward footprint along the trajectory

---

Quickstart

KISS-ICP odometry

from apairo.dataset.rellis import Rellis3DDataset
from apairo_preprocess import KissICPOdometry

Rellis3DDataset.run_preprocess(
    KissICPOdometry(voxel_size=1.0),
    "/data/Rellis-3D/00000",
)
# writes kissicp_poses/000000.npy, 000001.npy, ...

Traversability from semantic labels

from apairo.dataset.rellis import Rellis3DDataset
from apairo_preprocess import TraversabilityFromLabels

# Default traversable IDs for RELLIS-3D: {dirt, grass, asphalt, concrete, puddle, mud}
Rellis3DDataset.run_preprocess(
    TraversabilityFromLabels(),
    "/data/Rellis-3D/00000",
)
# writes trav_label/000000.npy, ...  (uint8: 1=traversable, 0=not)

Custom IDs for SemanticKITTI:

from apairo.dataset.semantic_kitti import SemanticKittiDataset
from apairo_preprocess import TraversabilityFromLabels

SemanticKittiDataset.run_preprocess(
    TraversabilityFromLabels(traversable_ids=frozenset({40, 44, 48, 49, 60, 72})),
    "/data/sequences/00",
)

Traversability ground truth from trajectory

Requires poses to be computed first (e.g. with KissICPOdometry).

import numpy as np
from apairo.dataset.goose import Goose3DDataset
from apairo_preprocess import KissICPOdometry, TraversabilityFromTrajectory

# Step 1 — odometry
Goose3DDataset.run_preprocess(
    KissICPOdometry(voxel_size=1.0),
    "/data/goose/seq_001",
)

# Step 2 — load poses and compute traversability ground truth
ds = Goose3DDataset("/data/goose/seq_001", keys=["kissicp_poses"])
poses = np.stack([ds[i].data["kissicp_poses"] for i in range(len(ds))])

Goose3DDataset.run_preprocess(
    TraversabilityFromTrajectory(poses, robot_radius=0.75, height_min=-0.3, height_max=0.5),
    "/data/goose/seq_001",
)
# writes trav_gt/000000.npy, ...  (uint8: 1=traversable, 0=not)

---

Examples

Ready-to-run scripts in examples/:

# KISS-ICP odometry on any supported dataset
python examples/kissicp_odometry.py /data/Rellis-3D/00000 --dataset rellis

# Traversability from semantic labels
python examples/traversability_from_labels.py /data/Rellis-3D/00000

# Traversability ground truth from trajectory (runs odometry first if needed)
python examples/traversability_from_trajectory.py /data/goose/seq_001

---

License

MIT