Home

Awesome

Value Iteration Networks

Code for NIPS 2016 paper:

Value Iteration Networks

Aviv Tamar, Yi Wu, Garrett Thomas, Sergey Levine, and Pieter Abbeel

UC Berkeley

Requires:

For generating the gridworld data and visualizing results, also requires:

To start: the scripts directory contains scripts for generating the data, and training the different models.

scripts/make_data_gridworld_nips.m generates the training data (random grid worlds). Alternatively, you can use the existing data files in the data folder (instead of generating them).

scripts/nips_gridworld_experiments_VIN.sh shows how to train the VIN models.

After training, a weights file (e.g., /results/grid28_VIN.pk) will be created. You can then run:

Related implementations:

Kent Sommer's implementation of VINs (including data generation) in python + pytorch

https://github.com/kentsommer/pytorch-value-iteration-networks

Abhishek Kumar's implementation of VINs in Tensor Flow

https://github.com/TheAbhiKumar/tensorflow-value-iteration-networks