Operational Questions for Policy-Gradient Control: Control Randomisation versus Reproducible Evaluation

Abstract

Research on policy-gradient control is represented by distinct lines of work, including Control Randomisation Approach for Policy Gradient and Application to Reinforcement Learning…, Continuous Parameter Control in Genetic Algorithms using Policy Gradient Reinforcement Learning, and Policy ensemble gradient for continuous control problems in deep reinforcement learning. The references are mapped across control randomisation, reproducible evaluation, and stochastic perturbations. Particular attention is paid to how evaluation protocols shape apparently conflicting conclusions and how those conclusions should be interpreted outside their original settings. The resulting evidence map identifies well-supported practices, unresolved tensions, and concrete priorities for comparative research. It emphasizes transparent assumptions, reproducible protocols, and evaluation measures that match the intended use.