The remarkable performance of Large Language Models (LLMs) has demonstrated great potential in planning capabilities. However, generating plans in the travel scenario remains challenging due to interdependent multiple constraints and the lack of effective feedback. Existing methods typically involve rule-based retrieval and limited resources, making it difficult to accurately meet user requirements and adhere to commonsense. We introduce MATP, a novel framework that simulates collaborative intelligence with multi-agent in travel planning, which perceives multiple constraints according to user requirements and autonomously decomposes tasks, then differentiates into multiple agent roles and rationally uses tools to collect relevant information for multi-round collaboration through providing adaptive suggestions and critical feedbacks, thereby generating feasible travel plans. Experiments on travel planning indicate that our method delivers superior performance compared to existing solutions.
Zeng et al. (Mon,) studied this question.