A method (300), and corresponding systems (400) and computer-readable mediums (426), for implementing a hierarchical multi-agent control system for an environment. A method (300) includes generating (306) an observation of an environment (104) by a first agent process (206) and sending (308) a first message (464) that includes the observation to a meta-agent process (202). The method includes receiving (312) a second message (464) that includes a goal (454), by the first agent process (206) and from the meta-agent process (202). The method includes evaluating (314) a plurality of actions (456), by the first agent process (206) and based on the goal (454), to determine a selected action (456). The method includes applying (316) the selected action (456) to the environment (104) by the first agent process (206).
展开▼