Fig. 1: Modular agentic planner (MAP).
From: A brain-inspired agentic architecture to improve planning with LLMs
The agent receives states from the environment and high-level goals. These are processed by a set of specialized LLM modules. The Task Decomposer receives the current state and a high-level goal and generates a series of subgoals. The Actor generates proposed actions given a state and a subgoal. The Monitor gates these proposed actions based on whether they violate certain constraints (e.g., task rules) and provides feedback to the Actor. The Predictor predicts the next state given the current state and a proposed action. The Evaluator is used to estimate the value of a predicted state. The Predictor and Evaluator are used together to perform tree search. The Orchestrator determines when each subgoal has been achieved, and when the final goal has been achieved, at which point the plan is emitted to the environment as a series of actions.
