Hierarchical Decentralized Deep Reinforcement Learning Architecture for a Simulated Four-Legged Agent

Legged locomotion is widespread in nature and has inspired the design of current robots. The controller these legged robots often realized as one centralized instance. However, nature, control movement happens a hierarchical decentralized fashion. Introducing biological principles into robotic systems motivated this work. We tackle question whether beneficial for present novel decentral, architecture to simulated agent. Three different tasks varying complexity are designed benchmark five architectures (centralized, decentralized, two combinations architectures). results demonstrate that decentralizing levels facilitates learning agent, ensures more energy efficient movements well robustness towards new unseen environments. Furthermore, comparison sheds light on importance modularity solve complex goal-directed tasks. provide an open-source code implementation our ( https://github.com/wzaielamri/hddrl ).