CVAE Exploration Planning Data

Year: 2022
DOI: 10.3929/ethz-c-000788332

Data accompanying the CVAE exploration planning project, providing local maps with supervision signals for conditional variational autoencoder and CNN models used in learning-based exploration.

Repository: cvae_exploration_planning • Preprint on ArXiv

Data Layout

• CVAE_dataset.npy: Local maps with best actions from a teacher planner (supervision for CVAE)
• CNN_dataset.npy: Local maps with random actions (CNN training set)
• test_worlds.zip: Test worlds
  • <environment>_<no>.p: Pickled test world
  • <environment>_<no>.txt: Initial pose [x, y, yaw]

Additional utilities for processing and ROS playback are in the panoptic_mapping repository.

Downloads

• Dataset landing page: https://doi.org/10.3929/ethz-c-000788332
• CVAE dataset (326.4 MB): https://doi.org/10.3929/ethz-c-000788332
• CNN dataset (1.0 GB): https://doi.org/10.3929/ethz-c-000788332
• Test worlds (56.5 KB): https://doi.org/10.3929/ethz-c-000788332

Citation

L. Schmid, C. Ni, et al.,
“Fast and Compute-efficient Sampling-based Local Exploration Planning via Distribution Learning”, IEEE Robotics and Automation Letters (RA-L), 2022.

Tags: exploration, planning, learning, mapping