Hierarchical Agent Architecture
A classic paradigm in reinforcement learning and robot control that splits the decision process across multiple time scales: a high-level policy handles long-horizon subgoal planning, while a low-level policy handles millisecond-level execution of concrete actions.
Timeline (last 90 days)
Using a large language model as a high-level planner paired with lightweight specialized executors is a general agent design pattern.
The hierarchical design (LLM for planning, lightweight models for execution) is becoming the mainstream paradigm in scenarios such as game AI, robot control and automated workflows.
All Facts (2)
Using a large language model as a high-level planner paired with lightweight specialized executors is a general agent design pattern.
50%UnverifiedThe hierarchical design (LLM for planning, lightweight models for execution) is becoming the mainstream paradigm in scenarios such as game AI, robot control and automated workflows.
50%