nuPlan 123D

nuPlan v1.1 in the 123D Apache Arrow format. 15910 logs, all modalities, 7.8 TiB. Splits: nuplan_train, nuplan_val, nuplan_test.

Layout

Path	Content
logs/{split}/{log_name}/*.arrow	1457 logs with sensors, one file per modality
logs_sensorless/{split}/NNNNNN.tar	14453 logs without sensors, 100 logs per uncompressed tar
maps/nuplan/	4 city maps
index.parquet	one row per log

Metadata modalities, present in every log, on the 10 Hz sync clock. Sensorless logs are packed into tars with entries {split}/{log_name}/{file}.arrow.

index.parquet columns: log_name, split, has_sensors, sensorless_tar (null for sensor logs), bytes.

Extract the tars into logs/ to get a single 123D tree:

123D/
├── logs/
│   ├── nuplan_test/
│   ├── nuplan_train/
│   └── nuplan_val/
│       └── 2021.06.07.11.59.52_veh-35_00008_00083/
│           ├── box_detections_se3.arrow
│           ├── camera.pcam_b0.arrow
│           ├── camera.pcam_f0.arrow
│           ├── camera.pcam_l0.arrow
│           ├── camera.pcam_l1.arrow
│           ├── camera.pcam_l2.arrow
│           ├── camera.pcam_r0.arrow
│           ├── camera.pcam_r1.arrow
│           ├── camera.pcam_r2.arrow
│           ├── custom.scenario.arrow
│           ├── ego_state_se3.arrow
│           ├── lidar.lidar_merged.arrow
│           ├── route_position.arrow
│           ├── sync.arrow
│           └── traffic_light_detections.arrow
└── maps/
    └── nuplan/
        ├── nuplan_sg-one-north.arrow
        ├── nuplan_us-ma-boston.arrow
        ├── nuplan_us-nv-las-vegas-strip.arrow
        └── nuplan_us-pa-pittsburgh-hazelwood.arrow

Conversion

Converted with py123d from the nuPlan v1.1 DBs and sensor archives.

• Sync at 10 Hz on the box detection timestamps, which equal the lidar timestamps.
• Camera images re-encoded as JPEG at OpenCV quality 75, native resolution.
• Lidar sweeps stored as laz with 2 cm xyz quantization, plus intensity, ring and lidar id. Lidar extrinsics are unknown in nuPlan and set to identity.
• route_position is derived from the ego odometry and records what it was computed from. Needs py123d >= 0.7.0.

License and citation

Derived from nuPlan; the nuPlan Dataset License applies, see LICENSE. If you use this dataset, please cite 123D and nuPlan.

@article{Dauner2026ARXIV,
  title={123D: Unifying Multi-Modal Autonomous Driving Data at Scale},
  author={Dauner, Daniel and Charraut, Valentin and Berle, Bastian and Li, Tianyu and Nguyen, Long and Wang, Jiabao and Jing, Changhui and Igl, Maximilian and Caesar, Holger and Ivanovic, Boris and Geiger, Andreas and Chitta, Kashyap},
  journal={arXiv preprint arXiv:2605.08084},
  year={2026}
}

@article{Caesar2021ARXIV,
  title={nuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles},
  author={Caesar, Holger and Kabzan, Juraj and Tan, Kok Seang and Fong, Whye Kit and Wolff, Eric and Lang, Alex and Fletcher, Luke and Beijbom, Oscar and Omari, Sammy},
  journal={arXiv preprint arXiv:2106.11810},
  year={2021}
}