Search papers, labs, and topics across Lattice.
This paper introduces a framework that uses a Mixture-of-Agents (MoA) of LLM-based experts to reduce human intervention in interactive task planning for robots. The framework generates Behavior Trees (BTs) to represent task logic hierarchically, enabling retry mechanisms and dynamic policy switching. Experiments on cocktail and smoothie-making tasks demonstrate a 27% reduction in human response requirements and successful long-horizon execution with adaptive policy switching and failure recovery.
LLMs can plan complex robot tasks with 27% less human input by using a mixture-of-agent system to answer questions and behavior trees to structure execution.
Interactive task planning with large language models (LLMs) enables robots to generate high-level action plans from natural language instructions. However, in long-horizon tasks, such approaches often require many questions, increasing user burden. Moreover, flat plan representations become difficult to manage as task complexity grows. We propose a framework that integrates Mixture-of-Agents (MoA)-based proxy answering into interactive planning and generates Behavior Trees (BTs) for structured long-term execution. The MoA consists of multiple LLM-based expert agents that answer general or domain-specific questions when possible, reducing unnecessary human intervention. The resulting BT hierarchically represents task logic and enables retry mechanisms and dynamic switching among multiple robot policies. Experiments on cocktail-making tasks show that the proposed method reduces human response requirements by approximately 27% while maintaining structural and semantic similarity to fully human-answered BTs. Real-robot experiments on a smoothie-making task further demonstrate successful long-horizon execution with adaptive policy switching and recovery from action failures. These results indicate that MoA-assisted interactive planning improves dialogue efficiency while preserving execution quality in real-world robotic tasks.
