Autonomous Planning
Learn to build autonomous agents that can plan and execute complex tasks by decomposing goals, estimating resources, and adapting to unexpected changes using classical planning, LLMs, and hybrid neuro-symbolic methods.
4 Steps
- 1
Classical Planning with PDDL: Define a simple planning problem using PDDL (Planning Domain Definition Language). This involves specifying the initial state, goal state, actions, and preconditions/effects of each action. For example, a robot moving boxes in a warehouse.
- 2
Solving PDDL Problems with a Planner: Use a classical planner (e.g., Fast Downward, pyperplan) to generate a plan for the PDDL problem defined in Step 1. Provide the planner with the domain and problem files to obtain a sequence of actions that achieve the goal.
- 3
LLM-Based Plan Generation: Use a Large Language Model (LLM) like GPT-3 or a fine-tuned model to generate plans directly from natural language descriptions of goals. Prompt the LLM with a clear task description and desired output format.
- 4
Hybrid Neuro-Symbolic Planning: Combine LLM-generated plans with classical planning for refinement and execution. Use the LLM to generate an initial high-level plan, then use a classical planner to fill in the details and ensure feasibility. Implement a mechanism for replanning when encountering failures.
Ready to run this action pack?
Activate your free AaaS account to access all packs, earn credits, and deploy agentic workflows.
Get Started Free →