PlanGEN: A Multi-Agent Framework for Generating Planning and Reasoning Trajectories for Complex Problem Solving
Mihir Parmar, Xin Liu, Palash Goyal, Yanfei Chen, Long Le, Swaroop Mishra, Hossein Mobahi, Jindong Gu, Zifeng Wang, Hootan Nakhost, Chitta Baral, Chen-Yu Lee, Tomas Pfister, Hamid Palangi
2025-02-28
Summary
This paper talks about PlanGEN, a new way to help AI systems solve complex planning problems more effectively. It's like giving AI a team of smart assistants to break down big problems, check its work, and choose the best method for each task.
What's the problem?
Current AI systems often struggle with complicated planning tasks because they can't always check if their plans will work or adjust their approach based on how hard each specific problem is. It's like trying to solve a bunch of different puzzles but using the same strategy for all of them, even when some puzzles are much harder than others.
What's the solution?
The researchers created PlanGEN, which uses three special AI agents working together. One agent figures out the rules and limits of each problem, another checks if the plans will actually work, and the third chooses the best problem-solving method based on how complicated the task is. PlanGEN also keeps improving its plans step by step, making sure they follow all the rules and work well.
Why it matters?
This matters because it could make AI much better at solving real-world problems that involve lots of planning, like scheduling complex projects or figuring out the best way to use resources. PlanGEN performed better than other top AI systems on several tough tests, which means it could help make AI more useful and reliable for all kinds of important tasks in business, science, and everyday life.
Abstract
Recent agent frameworks and inference-time algorithms often struggle with complex planning problems due to limitations in verifying generated plans or reasoning and varying complexity of instances within a single task. Many existing methods for these tasks either perform task-level verification without considering constraints or apply inference-time algorithms without adapting to instance-level complexity. To address these limitations, we propose PlanGEN, a model-agnostic and easily scalable agent framework with three key components: constraint, verification, and selection agents. Specifically, our approach proposes constraint-guided iterative verification to enhance performance of inference-time algorithms--Best of N, Tree-of-Thought, and REBASE. In PlanGEN framework, the selection agent optimizes algorithm choice based on instance complexity, ensuring better adaptability to complex planning problems. Experimental results demonstrate significant improvements over the strongest baseline across multiple benchmarks, achieving state-of-the-art results on NATURAL PLAN (sim8%uparrow), OlympiadBench (sim4%uparrow), DocFinQA (sim7%uparrow), and GPQA (sim1%uparrow). Our key finding highlights that constraint-guided iterative verification improves inference-time algorithms, and adaptive selection further boosts performance on complex planning and reasoning problems.