Deep Gaussian Process for Inverse Reinforcement Learning
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Binaryworld
deepGPIRL
DGP-IRL UAI 2017.pdf
LICENSE
README.md

README.md

DGP-IRL: Deep Gaussian Process for Inverse Reinforcement Learning

How to use the code

To run the algorithm, please first download the toolbox by Levine et. al. (2010): http://graphics.stanford.edu/projects/gpirl/irl_toolkit.zip

Then put the folder of deepGPIRL and Binaryworld in the same folder, and add the paths to these folders. You don't need to specify the parameters to start with. You can modify them in the file deepgpirldefaultparams.m .

Example to run the DGP-IRL algorithm on binary world benchmark: test_result_dgpirl = runtest('deepgpirl',struct(),'linearmdp','binaryworld',struct('n',12),struct('training_sample_lengths',12^2,'training_samples',8,'verbosity',1));

Reference

Ming Jin, Andreas Damianou, Pieter Abbeel, and Costas Spanos, "Inverse reinforcement learning via deep Gaussian Process", In Conference on Uncertainty in Artificial Intelligence (UAI 2017)