onlytailei / Value-Iteration-Networks-PyTorch

PyTorch implementation of the Value Iteration Networks (VIN) (NIPS '16 best paper)

Value Iteration Networks in PyTorch

Tamar, A., Wu, Y., Thomas, G., Levine, S., and Abbeel, P. Value Iteration Networks. Neural Information Processing Systems (NIPS) 2016

This repository contains an implementation of Value Iteration Networks (VIN) in PyTorch based on the original Theano implementation by the authors and the TensoFlow implementation by Abhishek Kumar.

VIN won the Best Paper Award at NIPS 2016.

Dependencies

• Python 2.7
• PyTorch
• SciPy >= 0.18.1 (to load the data)

Datasets

• The datasets is from the author's repository. This repository contains the 8x8 GridWorld dataset for convenience and its small size.
• utils.py and data.py are from Abhishek Kumar's repository

Training

python train_main.py

Several arguments can be set in train_main.py like learning rate. Please check train_main.py for details.

python train_main.py --lr 0.001

References

• Value Iteration Networks on arXiv
• Aviv Tamar's (author) implementation in Theano
• Abhishek Kumar's implementation in TensorFlow.