Bioinspired actor-critic algorithm for reinforcement learning interpretation with Levy-Brown hybrid exploration strategy

Currently, reinforcement learning, the interpretability of the algorithm is a challenge. The lack of interpretability limits the use of reinforcement learning limited when facing agents in the physical world. To improve the interpretability of reinforcement learning, this study proposes a Levy-Brown hybrid strategy to improve the working of the traditional Actor-Critic algorithm. The proposed strategy is bioinspired from the Brown motion and Levy motion in nature; therefore, it can explain the process of data acquisition in the learning process from biological principles. The main idea of this new strategy is to map the Gaussian strategy to the biological Brown motion, and introduce the biological Levy strategy to improve the exploration efficiency. By combining the two strategies, it effectively takes advantage of the Levy strategy to improve exploration speed and the Brown strategy to improve exploration stability. The experiments demonstrate the advantages of the proposed Levy-Brown hybrid strategy, which effectively make best use of the advantages and overcomes the disadvantages of the two strategies.