Value activation for bias alleviation: Generalized-activated deep double deterministic policy gradients

•We propose a novel generalized-activated weighting operator for bias alleviation in deep reinforcement learning.•We show theoretically and experimentally that generalized-activated weighting operator helps alleviate both underestimation bias and overestimation bias.•We find that simple activation functions are enough for amazing performance without any tricks and special design for activation function.