Robust Quantum Control using Reinforcement Learning from Demonstration

This project explores robust and sample-efficient quantum control using Reinforcement Learning from Demonstration (RLfD). The goal is to overcome model bias in quantum control tasks and reduce the training time typically required by model-free reinforcement learning (RL) algorithms.

We combine classical control pulse demonstrations (generated from system models) with RL to optimize quantum gate calibration and non-classical state preparation. Our approach enables sample-efficient training for large-scale control problems, including those involving over 1000 discretized pulse segments.

Key Features

• Implements RLfD to accelerate learning by using model-generated demonstrations.
• Compatible with the Gymnasium API.
• Built on top of Stable-Baselines3 and Pytorch.

Pre-Request

Firstly, install the dependencies in pyproject.toml.

Then

# install the environments of quantum control problems
cd gym-quantum
pip install -e .
# You can build your own control problems.

Misc

This project may contain bugs. Feel free to open issues or contribute!