Future of AI and Empowering Reinforcement Learning with Meta-Critic Networks and GAE in a Human-Centered Framework

The widespread application of artificial intelligence (AI) has significantly enhanced people’s quality of life, making AI-driven technologies an integral part of our daily lives. This trend indicates that human-centered computing will inevitably become a focal point in the development of next-generation AI technologies. Among the most critical AI algorithms, reinforcement learning (RL) stands out as a particularly hot area of development. RL enables agents to interact with their environments, receive rewards, and optimize their decision-making strategies through trial and error. RL algorithms are essential in fields such as urban path planning and navigation, autonomous driving, humanoid robotics, and urban network edge computing. Despite their potential, RL algorithms face challenges including low sample efficiency, weak generalization, and lengthy computational cycles. Meta-RL algorithms are regarded by the academic community as a promising solution to these issues. This paper enhances meta-critic network by integrating with the actor-critic algorithm and the generalized advantage estimation (GAE) module. We propose the meta-critic reinforcement learning by using GAE algorithm, which achieves a 39.9% performance improvement compared to the baseline algorithm. After extensive training in numerous similar environments, the proposed algorithm can rapidly converge and achieve optimal results in new environments. Extensive experiments demonstrate that the proposed algorithm excels in both performance and stability.