nuPlan 123D

nuPlan v1.1 in the 123D Apache Arrow format: all 15910 logs with every modality, 7.8 TiB.

• logs/{nuplan_train,nuplan_val,nuplan_test}/{log_name}/: the 1457 sensor-bearing logs as plain files, one per modality: camera.pcam_{f0,l0,r0,l1,r1,l2,r2,b0}, lidar.lidar_merged and the metadata (sync, ego_state_se3, box_detections_se3, traffic_light_detections, route_position, custom.scenario), all on the 10 Hz sync clock. route_position is derived from the ego odometry and records what it was computed from (needs py123d >= 0.7).
• logs_sensorless/{split}/NNNNNN.tar: the 14453 logs without sensors, metadata only, 100 logs per uncompressed tar. Entries are {split}/{log_name}/{file}.arrow, so extracting into logs/ yields the flat 123D tree.
• maps/nuplan/: the 4 city maps.
• index.parquet: one row per log with log_name, split, has_sensors, sensorless_tar (null for sensor logs) and bytes.

Usage

pip install py123d
export PY123D_DATA_ROOT=/path/to/nuplan

# sensorless logs + maps (0.4 TiB): planning without sensors
hf download kesai-labs/nuplan --repo-type dataset --local-dir $PY123D_DATA_ROOT \
    --include "logs_sensorless/*" --include "maps/*" --include "index.parquet"
for tar_file in $PY123D_DATA_ROOT/logs_sensorless/*/*.tar; do
    tar -xf "$tar_file" -C $PY123D_DATA_ROOT/logs
done

# sensor logs, metadata only (50 GiB)
hf download kesai-labs/nuplan --repo-type dataset --local-dir $PY123D_DATA_ROOT \
    --include "logs/*" --exclude "logs/*/*/camera.*" --exclude "logs/*/*/lidar.*"

# then any modality on top, e.g. the front camera (0.7 TiB)
hf download kesai-labs/nuplan --repo-type dataset --local-dir $PY123D_DATA_ROOT \
    --include "logs/*/*/camera.pcam_f0.arrow"

See the py123d documentation for the Scene and Map API.

License

Derived from nuPlan; the nuPlan Dataset License applies, see LICENSE.