Creating valid warehouses

A valid warehouse layout is a special kind of layout that follows certain rules. For more information, refer to https://arxiv.org/abs/2305.06436.

This repo contains code that creates those warehouse environments using diffusion models. It uses huggingface APIs to train a diffusion model from scratch.

using_diffusers_on_numpy_arrays.ipynb can be run by itself without any dependencies. The code also generates its own training data and trains the model.

Depth first approach to create the training data

dfs.py is the logic that creates the training data. The warehouse layout needs to follow certain physical rules and DFS was used to acheive this. using_diffusers_on_numpy_arrays.ipynb uses a fixed dimension warehouse, to change the dimension, the DFS logic in it needs to be changed.

Chain of thought approach

A Chain of thought approach was also tried to achieve the warehouse layout that follows the constraints that were preveiously mentioned. OpenAI's API was used for this approach. Although the results were not satisfactory, the code can be found at chain_of_thought.py. It will require an OpenAI API key.