EUBRL: Epistemic Uncertainty Directed Bayesian Reinforcement Learning

An implementation of EUBRL, a Bayesian RL algorithm that leverages epistemic guidance to achieve principled exploration.

Installation

python install.py

Usage

Examples are given for running EUBRL. Any other algorithms in the same folder can be run in a similar manner.

Chain

bash scripts/Chain/EUBRL.sh

Loop

bash scripts/Loop/EUBRL.sh

DeepSea

Stochastic

bash scripts/DeepSea-Stochastic/EUBRL.sh

Deterministic

bash scripts/DeepSea-Deterministic/EUBRL.sh

LazyChain

Stochastic

bash scripts/LazyChain-Stochastic/EUBRL.sh

Deterministic

bash scripts/LazyChain-Deterministic/EUBRL.sh

Acknowledgment

Our code is based on open-source repository BayesRL.

License

This project is licensed under the GNU General Public License v3.0 (GPLv3).