HELM: Human-Preferred Exploration with Language Models

TL;DR

HELM addresses adaptive human-preferred exploration in unknown environments by integrating a pre-trained LLM with a perception-driven belief graph. It builds a collision-free viewpoint graph and a sparse information graph, translates human directives into task-specific prompts, and uses an LLM-based planner to select waypoint sequences with closed-loop control. Across simulations and Gazebo tests, HELM achieves competitive exploration efficiency and demonstrates superior adaptability when guided by natural-language preferences, all without retraining. The work showcases the practical potential of LLM-guided planning for real-world robotic exploration and sets the stage for broader human-robot collaboration in navigation and mapping.

Abstract

In autonomous exploration tasks, robots are required to explore and map unknown environments while efficiently planning in dynamic and uncertain conditions. Given the significant variability of environments, human operators often have specific preference requirements for exploration, such as prioritizing certain areas or optimizing for different aspects of efficiency. However, existing methods struggle to accommodate these human preferences adaptively, often requiring extensive parameter tuning or network retraining. With the recent advancements in Large Language Models (LLMs), which have been widely applied to text-based planning and complex reasoning, their potential for enhancing autonomous exploration is becoming increasingly promising. Motivated by this, we propose an LLM-based human-preferred exploration framework that seamlessly integrates a mobile robot system with LLMs. By leveraging the reasoning and adaptability of LLMs, our approach enables intuitive and flexible preference control through natural language while maintaining a task success rate comparable to state-of-the-art traditional methods. Experimental results demonstrate that our framework effectively bridges the gap between human intent and policy preference in autonomous exploration, offering a more user-friendly and adaptable solution for real-world robotic applications.

Figures (6)

• Figure 1: Our human-preferred exploration framework uses a pre-trained large-language model (LLM), which can follow human preferences to perform long-term planning based on onboard sensor data.
• Figure 2: HELM processes onboard sensor data and combines it with human preferences and real-time intermediate data to construct human-preferred queries for decision-making. The Task Prompts Questioner can either use a predefined template or the LLM Agent for automatic generation.
• Figure 3: sparse information graph.
• Figure 4: Exploration paths comparisons in a large-scale $130m \times 100m$ indoor office simulation.
• Figure 5: Different human preferences lead to different exploration strategies. The upper part of the image shows the robot's trajectory and the partial map obtained during the exploration process. The lower part shows the ground truth, with yellow representing the explored areas and purple representing the unexplored areas.