Monte Carlo Beam Search for Actor-Critic Reinforcement Learning in Continuous Control

Hazim Alzorgan, Abolfazl Razi

Actor-critic methods, like Twin Delayed Deep Deterministic Policy Gradient (TD3), depend on basic noise-based exploration, which can result in less than optimal policy convergence.…