apairo-rr

Rerun visualisation layer for apairo datasets.

Logs LiDAR point clouds, semantic labels, and robot trajectories to the Rerun viewer. Supports multiple side-by-side pipelines, sequence-aware navigation, and any apairo dataset (RELLIS-3D, SemanticKITTI, GOOSE, TartanKitti…).

---

Installation

pip install -e .

Requires apairo and apairo_preprocess (resolved from local paths in pyproject.toml).

---

Quickstart

import apairo
import apairo_rr
from apairo_rr import Pipeline

ds = apairo.Rellis3DDataset("/data/RELLIS", keys=["lidar", "labels"])

apairo_rr.view(
    ds,
    label_cfgs=[apairo_rr.load_label_config("rellis")],
    pipelines=[Pipeline("Semantic GT")],
)

---

API

view(dataset, *, ...)

Main entry point. Logs a dataset to the Rerun viewer.

Parameter	Type	Description
dataset	apairo dataset	Any dataset supporting dataset[idx] → Sample
label_cfg	dict | None	Label config applied to all pipelines
label_cfgs	list[dict | None]	Per-pipeline label configs (overrides label_cfg)
poses	list[np.ndarray]	4×4 pose matrices — logs a trajectory overlay
pipelines	list[Pipeline]	Processing pipelines, one 3D view each
point_key	str	Sample key for the point cloud (default "lidar")
label_key	str | None	Sample key for semantic labels (default "labels")
frames	Iterable[int] | None	Frame indices to log (default: all)
spawn	bool	Spawn the Rerun viewer automatically (default True)

If the dataset exposes sequence_ids (all ProfiledDataset subclasses), a Sequence panel appears in the viewer and updates as you scrub the timeline.

Pipeline(name, steps=[])

Named sequence of per-frame transforms.

Pipeline("Raw")
Pipeline("Range filter", [range_filter])
Pipeline("Trav — labels", [range_filter, TraversabilityFromLabels()])

Each step must have signature (pts, labels) -> (pts, labels) or (pts, labels, frame_idx=...) -> (pts, labels). apairo_preprocess FramePreprocessor objects are accepted directly.

load_label_config(name)

Load a built-in label config by name.

cfg = apairo_rr.load_label_config("rellis")          # RELLIS-3D (20 classes)
cfg = apairo_rr.load_label_config("semantic_kitti")  # SemanticKITTI (28 classes)
cfg = apairo_rr.load_label_config("goose")           # GOOSE (64 classes)

Returns a dict with color_map and semantic_map keys, compatible with view(label_cfgs=...).

---

Examples

RELLIS-3D — three-way traversability comparison

# All sequences
python examples/view_rellis_traversability.py --root ~/data/rellis

# Single sequence, every 5th frame
python examples/view_rellis_traversability.py --sequence 00000 --every 5

# Custom robot radius for trajectory-based traversability
python examples/view_rellis_traversability.py --sequence 00001 --radius 0.8

CLI options

Flag	Default	Description
--root	~/data/rellis	Dataset root directory
--sequence	(all)	Sequence ID to visualise (00000, 00001, …)
--every N	1	Log every Nth frame
--idx N	0	Start at frame N (within the selected sequence)
--radius R	1.0	Robot radius for TraversabilityFromTrajectory

---

Sequence navigation

When --sequence is omitted, all sequences are concatenated on the same timeline. Scrub the Rerun timeline to move between frames — the Sequence panel at the top of the viewer shows the current sequence ID.

To jump directly to a sequence, relaunch with --sequence <id>. Available IDs are printed at startup:

  1832 scans — sequences: ['00000', '00001', '00002', '00003', '00004']

---

Adding a custom dataset

Any object supporting __getitem__(int) -> Sample and __len__ works with view(). For sequence-aware navigation, the dataset must expose:

dataset.sequence_ids          # list[str]
dataset.sequence(seq_id)      # SequenceView with ._indices (global frame indices)

All apairo.ProfiledDataset subclasses (RELLIS, SemanticKITTI, GOOSE) implement this automatically.