DUPLEX: Agentic Dual-System Planning via LLM-Driven Information Extraction

1. 1Identify the limitations of LLMs in long-horizon task planning, such as hallucination and logical inconsistency
2. 2Design a dual-system architecture that separates information extraction from plan synthesis
3. 3Implement schema-guided information extraction using LLMs to provide semantic flexibility
4. 4Integrate the extracted information with symbolic planning to ensure rigorous plan synthesis

This research benefits AI engineers and roboticists working on task planning systems, as it provides a more reliable approach to planning in complex environments

💡 Combining LLMs with symbolic planning can improve the reliability of task planning systems

🤖 DUPLEX: A new neuro-symbolic architecture for reliable long-horizon task planning #LLMs #AI