I created this enviorment for the public to easily able to implement any Reinforcement Learning algorithm for Optimisation of Qubit pulses. My main motivation for this work is this paper https://arxiv.org/pdf/2105.01079.pdf. This enviorment hopes to provide an easy playground for creating policy networks for the Qubit Pulses without worrying about the Physics involved.

Do one of the following in the source directory (preferably in a Python virtual environment set up for Qiskit Dynamics)

• ./setup.py install
• pip3 install .

import gym
import Rx_Env
env = gym.make('Rx_Env/RxEnv-v0')

Right now the enviornment returns

1. observation: A dictionary of Start, Current, Target State
2. reward: Reward directly based on Fidelity, some bonus is given for fidelity greater than 0.999
3. done: To stop the episode after a set number of evolutions(20) achieved
4. truncated: To stop in between for some reason.
5. fidelity info: Returns a Dictionary of current Fidelity

Please raise an issue/Pr if you have any suggetions or would like to contribute. Ideally I would like for this library to contain a default Policy Network good enough to give you the ideal pulse for implementatiion of any gate given its paramenters and a hardware without requiring any Physics or RL knowledge. I am currently working on creating such policy network for X gate. If I can get it working efficiently enough I shall move to creating a unviversal Enviornment/Policy Network.