EvoPlan: Evolutionary Neuro-Symbolic Robot Planning with Spatio-Temporal Guarantees

2026-07-07

AI Summary

This work addresses the limitations of existing large language model (LLM)-driven robotic planning, which often lacks executability and safety guarantees, while traditional symbolic planners struggle to leverage contextual information and require fully formalized domain descriptions. The authors propose a neuro-symbolic framework that integrates an on-device LLM with an evolutionary PDDL planner, uniquely combining evolutionary search and counterfactual perturbation generation to automatically extract Signal Temporal Logic (STL) constraints from single-class demonstrations. This approach establishes a closed-loop plan-verify-execute architecture that preserves the flexibility of natural language while providing formal safety assurances. Evaluated on benchmarks including Bench2Drive, HA-VLN-CE, and ALFWorld, the method achieves higher planning success rates than strong baselines and demonstrates robustness even under mismatches between action and goal vocabularies.

Abstract

LLM-based robot planners are fluent but cannot guarantee that their plans are executable or safe. Classical PDDL planners can guarantee these properties, but only after the problem is fully specified, and they make poor use of an LLM's ability to read context and repair plans. This paper presents a neuro-symbolic framework with three parts. All LLM calls use a locally-hosted open-weight model, so the pipeline can be deployed on-robot with no cloud dependency. First, an offline procedure that mines a single global Signal Temporal Logic (STL) constraint on mobility from demonstration data. The procedure recovers codified rules (e.g., stopping at red lights, mined from nuPlan driving logs) or population preferences (e.g., social-navigation comfort, mined from SCAND teleoperation), depending on what the demonstrations encode. Because the demonstrations are a one-class signal, we generate the missing negatives with counterfactual perturbations and an LLM violation generator and then fit the constraint by evolutionary search. We use the mined constraint to shield a vision-language driving policy on Bench2Drive and two discrete-action navigation policies on HA-VLN-CE. Second, an evolutionary PDDL planner: an LLM proposes and repairs plans, programmatic validators decide which ones survive, and the validated portion of the plan grows over iterations. We test the planner on the open-world ALFWorld Text benchmark, where it beats strong baselines and stays robust when the goal vocabulary does not match the action-model vocabulary. Third, a constrained execution loop: the planner's plan is compiled into waypoints, the waypoints are checked against the mined constraint, and the planner re-plans on a violation. We illustrate the full pipeline via demonstrations using the Gazebo simulator.

Problem

• robot planning
• safety guarantees
• neuro-symbolic
• executable plans
• large language models

Innovation

• neuro-symbolic planning
• Signal Temporal Logic (STL)
• evolutionary search
• LLM-based planning
• constrained execution