Multi-stage AI planning framework improves complex task accuracy
GRASP: Generating, Revising, and Assessing for Strategic Planning with Agentic AI
Artificial IntelligenceComputation and LanguageMachine LearningMultiagent Systems
Summary
Solving complicated tasks with AI language models often becomes less reliable as the task grows harder. The authors introduce GRASP, a new step-by-step planning approach that breaks down the process into generating broad plans, revising options in smaller chunks, and checking the plan's quality carefully. This method helps AI models make better and more consistent plans, even when juggling many tasks at once. The authors tested GRASP on different challenges and found it outperforms previous methods by a notable margin.
What this means in practice
- •For calendar application developers: Improve scheduling assistants to create more accurate and reliable natural language event plans for complex calendar scenarios.
- •For workflow automation engineers: Design robust multi-step workflow planners that maintain performance across diverse and simultaneous tasks by isolating context and revising strategies.
- •For customer support chatbot builders: Create chatbots capable of handling complex multi-turn conversations by generating and verifying strategic dialogue plans.$Commercial implications: Enables selling advanced AI customer support products that reliably manage complex conversations beyond current capabilities.
Authors
Arunabh Srivastava, Mohammad A., Khojastepour, Srimat Chakradhar, Sennur Ulukus
Abstract
Large Language Models (LLMs) typically exhibit a performance profile where reliability degrades as task complexity increases. We address the challenge of generating high-quality natural language executable plans for complex tasks by introducing $\textbf{GRASP}$, a strategy-aware, multi-stage planning framework. GRASP decouples the planning pipeline across specialized, context-isolated modules: it pre-compiles global macro-guidelines (GenPlan), explores alternative localized strategies within isolated context windows (RevPlan), and independently evaluates trajectories using a multi-criteria discriminator (VerPlan). Empirical evaluations show that GRASP consistently establishes a new state-of-the-art frontier across diverse datasets, yielding substantial accuracy gains over direct LLM planners on Natural Plan Calendar Scheduling ($\sim$12.4$\%$$\uparrow$), ZebraLogic ($\sim$30.8$\%$$\uparrow$), and SciBench Math. Crucially, under multi-task scaling-where standard planners suffer immediate performance collapse-GRASP completely flattens the multi-task degradation penalty. In interleaved dual-task environments, GRASP achieves an absolute accuracy gain of up to 16.7$\%$ over direct LLM planners. Furthermore, by isolating context and enforcing strict macro-regularization, GRASP outperforms frontier reasoning models (such as GPT-5-mini) by a margin of 14.5$\%$.