GitHub - jannerm/diffuser at kuka (original) (raw)

Planning with Diffusion for Flexible Behavior Synthesis, in Pytorch

This is the implementation of the robotics experiments for Planning with Diffusion for Flexible Behavior Synthesis in Pytorch.

Usage

First, install and extract the dataset for training and pretrained models from this URL in the root directory of the repo.

To train the unconditional diffusion model on the block stacking task, you can use the following command:

You may evaluate the diffusion model on unconditional stacking with

python scripts/unconditional_kuka_planning_eval.py

or conditional stacking with

python scripts/conditional_kuka_planning_eval.py

The rewards are not normalized -- you need to divide numbers by 3.0 to get numbers reported in the paper.

Citations

@inproceedings{janner2022diffuser, title = {Planning with Diffusion for Flexible Behavior Synthesis}, author = {Michael Janner and Yilun Du and Joshua B. Tenenbaum and Sergey Levine}, booktitle = {International Conference on Machine Learning}, year = {2022}, }

Acknowledgements

The diffusion model implementation is based on Phil Wang's denoising-diffusion-pytorch repo. The organization of this repo and remote launcher is based on the trajectory-transformer repo.