Koopman-Assisted RL

Two novel KARL algorithms — Soft Koopman Value Iteration (SKVI) and Soft Actor Koopman-Critic (SAKC) — leverage the Koopman operator to linearize nonlinear dynamics for improved sample efficiency.

Control-Oriented Environments

Four benchmark environments rooted in the control literature: Linear System, Fluid Flow, Lorenz attractor, and Double-Well potential. Each environment targets a distinct class of dynamical behavior.

Modular & Reproducible

Built on CleanRL-style implementations with typed argument parsers, Optuna-based hyperparameter search, and pre-optimized configurations — making every experiment reproducible and easy to extend.