LLM-Flax : Generalizable Robotic Task Planning via Neuro-Symbolic Approaches with Large Language Models

TL;DR

LLM-Flax is a three-stage neuro-symbolic framework that automates domain rule generation, failure recovery, and object importance scoring using large language models, significantly reducing manual effort in robotic task planning.

Contribution

It introduces a fully automated, LLM-driven approach for domain rule creation, failure handling, and object scoring, eliminating manual rule authoring and training data needs.

Findings

Achieves higher success rates than manual baselines across multiple benchmarks.

Successfully performs zero-shot object importance scoring without training data.

Identifies context-window limitations at scale as a key challenge for future work.

Abstract

Deploying a neuro-symbolic task planner on a new domain today requires significant manual effort: a domain expert must author relaxation and complementary rules, and hundreds of training problems must be solved to supervise a Graph Neural Network (GNN) object scorer. We propose LLM-Flax, a three-stage framework that eliminates all three sources of manual effort using a locally hosted LLM given only a PDDL domain file. Stage 1 automatically generates relaxation and complementary rules via structured prompting with format validation and self-correction. Stage 2 introduces LLM-guided failure recovery with a feasibility-gated budget policy that explicitly reserves API latency cost before each LLM call, preventing the downstream relaxation fallback from being starved. Stage 3 replaces the domain-trained GNN entirely with zero-shot LLM object importance scoring, requiring no training data. We evaluate all three stages on the MazeNamo benchmark across 10x10, 12x12, and 15x15 grids (8 benchmarks total). LLM-Flax achieves average SR 0.945 versus the manual baseline's 0.828 (+0.117), matching or outperforming manual rules on every one of the eight benchmarks. On 12x12 Expert, LLM-Flax attains SR 0.733 where the manual planner fails entirely (SR 0.000); on 15x15 Hard, it achieves SR 1.000 versus Manual's 0.900. Stage 3 demonstrates feasibility (SR 0.720 on 12x12 Hard with no training data) but faces a context-window bottleneck at scale, pointing to the primary open challenge for future work.