Part of our code base for combining a/rqmc with pg methods.

Files of interest:

• main.py: main entry point for learning (not only LQR, but for all environments and methods).
• rqmc_distributions/normal_rqmc.py and rqmc_distributions/uniform_rqmc.py: define PyTorch Distributions classes for RQMC sampling, built on top of SSJ (also see rqmc_distributions/ssj_sobol.py which defines the SSJ Java → Python wrapper).
• envs/lqr.py: implementation of the LQR environment, which is typically instantiated as follows:

lqr = LQR(
    N=6,
        M=8,
        init_scale=3.0,
        max_steps=20,
        Sigma_s_kappa=1.0,
        Q_kappa=3.0,
        P_kappa=3.0,
        A_norm=1.0,
        B_norm=1.0,
        Sigma_s_scale=1.0,
        random_init=False,
        lims=100,
)