Abstract
MinMaxMin -learning is a novel optimistic Actor-Critic algorithm that addresses the problem of overestimation bias (-estimations are overestimating the real -values) inherent in conservative RL algorithms. Its core formula relies on the disagreement among -networks in the form of the min-batch MaxMin -networks distance which is added to the -target and used as the priority experience replay sampling-rule. We implement MinMaxMin on top of TD3 and TD7, subjecting it to rigorous testing against state-of-the-art continuous-space algorithms-DDPG, TD3, and TD7-across popular MuJoCo and Bullet environments. The results show a consistent performance improvement of MinMaxMin over DDPG, TD3, and TD7 across all tested tasks.