README.md

metadata

license: cc-by-4.0

Official point cloud dataset for the PushT manipulation task used in Does Latent Planning Survive Point Clouds? Action-Conditioned JEPA World Models for Geometric Observations and Goals (arXiv:2608.29434). Format, download and usage: github.com/fafraob/point-lewm (docs/data.md).

Versions

• v2 (main, October 2026): the pixels column holds the sensor's own view, a 224 x 224 render from the range sensor's camera pose, and the lidar column was recast with a corrected state replay. The v1 conversion restored the T block's position before its angle, which displaced the T in the scans wherever it is rotated; state, pixels and the actions were not affected.
• v1 (--revision v1, September 2026): the release the paper's point-cloud checkpoints were trained on, with the original scene-camera images and the displaced-T scans.

hf download fafraob/point-cloud-pusht pusht.lance.tar.zst --repo-type dataset --local-dir $PLWM_DATA_ROOT                 # v2
hf download fafraob/point-cloud-pusht pusht.lance.tar.zst --repo-type dataset --local-dir $PLWM_DATA_ROOT --revision v1   # v1