Crossway Diffusion: Improving Diffusion-based Visuomotor Policy via Self-supervised Learning

This is the checkpoint release of the ICRA'24 paper Crossway Diffusion: Improving Diffusion-based Visuomotor Policy via Self-supervised Learning.

This repo contains:

Pretrained models (500 epochs): ./checkpoints/crossway_diffusion-<task>-s<seed>.ckpt
SHA1 checksums: ./checkpoints/crossway_diffusion-<task>-s<seed>.ckpt.sha1
Evaluation results (1000 episodes): ./checkpoints/crossway_diffusion-<task>-s<seed>.ckpt.json

Available tasks: can_mh, can_ph, lift_mh, lift_ph, square_mh, square_ph, transport_mh, transport_ph, tool_hang_ph, pusht

Available seeds: 42, 43, 44

Please replace <task> with a specific task name and replace <seed> with 42, 43, or 44 to navigate the repo. For example, the path of pretrained model for task Push-T trained on seed 42 is ./checkpoints/crossway_diffusion-pusht-s42.ckpt

For the training and evaluation code, please visit our github repo.

Watch 🎬 video presentation!

For questions, please raise a github issue.