Agent-State Construction with Auxiliary Inputs

Reinforcement learning with auxiliary inputs.

Code for the corresponding paper published at TMLR.

Paper

Installation

Simply install everything in requirements.txt.

Running

All experiments are run through the main.py file. Check out the arguments file in unc/args.py for a list of all possible hyperparameter configurations.

Experiments

Experiments are defined in the hyperparameter files located in scripts/hparams.

Environments

Environments are set up such that you just need to specify the environment string as an argument (see unc.args for more details).

Changes to this base environment are mostly gym.Wrappers around this environment, in unc.envs.wrappers.