Language to Plan Robot Task Planner with LLM and PDDL

Can a small language model turn a spoken-style order into a correct robot plan? This project builds a language-to-plan task planner in the LLM+P style: a LangGraph planner agent reads the scene from a robot executor over the Agent2Agent (A2A) protocol, lets qwen2.5:3b translate the instruction into a PDDL goal, checks the goal, plans with pyperplan, validates the plan with an own STRIPS validator and repairs the goal once if needed; the A2A robot executor validates the plan again and executes it on a simulated arm in PyBullet. It is compared with the gold goal + planner, a rule parser, a random walk, the LLM writing the plan directly and an ablation without repair on PlanBench Blocksworld (50 instances x 2 phrasings). qwen2.5:3b was served by Ollama on a second laptop (RTX 2050 4 GB, 32 GB RAM) over Wi-Fi; the code ran on a laptop with RTX 2050 4 GB (Ryzen 7, 8 GB RAM).

NOTE: Research demo: plans are executed only on a simulated arm in PyBullet; the executor never moves a real robot and never runs a plan that its own validator rejects.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG11 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.