Quadrupped-Legged Robot Movement Plan Generation using Large Language Model

TL;DR

The paper tackles making quadruped navigation accessible through natural-language instructions by offloading LLM-based planning to an external server while maintaining real-time ROS-based local navigation. It introduces a distributed hardware/software stack on the DeepRobotics Jueying Lite 3, grounding Indonesian-language prompts into a $W_i$-driven movement plan via HDL-Localization for mapping and a structured JSON sequence executed by move_base. A key contribution is the prompt design for Vertex AI Gemini that yields a strict JSON action list describing navigations between semantic waypoints $W_i$ with global coordinates $(x,y,z)$, enabling robust indoor navigation. Experimental results across single-room, multi-room, and cross-zone tasks achieve aggregate success rates exceeding 90%, demonstrating feasibility for real-world deployment, with noted opportunities to improve multi-room navigation and plan-enhanced perception; future work includes retrieval-augmented generation and incorporating Visual Language Models for scene understanding.

Abstract

Traditional control interfaces for quadruped robots often impose a high barrier to entry, requiring specialized technical knowledge for effective operation. To address this, this paper presents a novel control framework that integrates Large Language Models (LLMs) to enable intuitive, natural language-based navigation. We propose a distributed architecture where high-level instruction processing is offloaded to an external server to overcome the onboard computational constraints of the DeepRobotics Jueying Lite 3 platform. The system grounds LLM-generated plans into executable ROS navigation commands using real-time sensor fusion (LiDAR, IMU, and Odometry). Experimental validation was conducted in a structured indoor environment across four distinct scenarios, ranging from single-room tasks to complex cross-zone navigation. The results demonstrate the system's robustness, achieving an aggregate success rate of over 90\% across all scenarios, validating the feasibility of offloaded LLM-based planning for autonomous quadruped deployment in real-world settings.

Figures (5)

• Figure 1: Movement Plan generation system architecture
• Figure 2: Point of interest in the main hall
• Figure 3: Website interface for natural language input
• Figure 7: Average Task Completion Time by Scenario
• Figure 8: Success Rate by Scenario