Hybrid Agentic AI-FSM Framework for Instruction-Based Industrial Manipulation Tasks
Sungmoon Joo, Ikjune Kim
AI summary
Problem
Translating natural language procedures into executable robot programs remains labor-intensive, while current LLM-based control models lack the determinism, verifiability, and safety guarantees required for industrial environments.
Approach
The framework uses an LLM to parse instructions into a structured task plan offline, which is then executed by a deterministic Finite State Machine. A multi-stage validation pipeline and human-in-the-loop exception handling ensure operational safety and reliability.
Key results
- Hierarchical schema-driven task planning from natural language
- Multi-stage validation and simulation dry-run pipeline
- RAG-based exception handling with mandatory human approval
- Independent safety actor layer enforcing real-time physical constraints
Why it matters
It provides a practical, safety-compliant architecture for deploying flexible AI in high-risk industrial automation, bridging the gap between LLM reasoning and traditional robotic control standards.
Abstract
This paper proposes a hybrid Agentic AI–FSM framework for robust natural-language-driven automation in safety-critical industrial robotics applications. Although natural-language procedures are commonplace in manufacturing, translating them into reliable robot programs remains labor-intensive. While Large Language Models (LLMs) offer strong parsing and planning capabilities, their inherent non-determinism and susceptibility to hallucinations preclude their direct use for robot control. To bridge this gap, our architecture employs an LLM-based planning agent to translate instructions offline into a structured task plan. Execution is then delegated to a deterministic Finite State Machine (FSM)-style execution engine to ensure reliability. Safety is further guaranteed by a multi-stage validation–simulation pipeline that verifies schema compliance and operational constraints through dry runs prior to deployment. For runtime anomalies, a RAG-enhanced Exception Handling Agent proposes recovery options, which are strictly mediated through a human-in-the-loop (HIL) interface for operator approval. Finally, a rule-based Safety Agent enforces physical constraints and provides an independent protection layer.