Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LogisticsVLN: Vision-Language Navigation For Low-Altitude Terminal Delivery Based on Agentic UAVs

Xinyuan Zhang, Yonglin Tian, Fei Lin, Yue Liu, Jing Ma, Kornélia Sára Szatmáry, Fei-Yue Wang

TL;DR

LogisticsVLN tackles the problem of precise, window-level terminal delivery using autonomous UAVs by integrating lightweight multimodal language and vision models into a modular VLN pipeline. The approach grounds natural language requests into target floors and windows via a Floor Count VLM and a target descriptor through a Request Understanding module, then performs object grounding and safe exploration with a Depth Assistant and a Viewpoint Selection strategy. A dedicated Visual-Language Delivery (VLD) dataset in CARLA, plus ablation analyses, demonstrate the system’s feasibility and identify bottlenecks in floor localization and object recognition across different VLMs. The work advances practical aerial delivery by providing a deployable, training-free framework and a benchmark for evaluating VLM-driven aerial delivery tasks in unseen environments.

Abstract

The growing demand for intelligent logistics, particularly fine-grained terminal delivery, underscores the need for autonomous UAV (Unmanned Aerial Vehicle)-based delivery systems. However, most existing last-mile delivery studies rely on ground robots, while current UAV-based Vision-Language Navigation (VLN) tasks primarily focus on coarse-grained, long-range goals, making them unsuitable for precise terminal delivery. To bridge this gap, we propose LogisticsVLN, a scalable aerial delivery system built on multimodal large language models (MLLMs) for autonomous terminal delivery. LogisticsVLN integrates lightweight Large Language Models (LLMs) and Visual-Language Models (VLMs) in a modular pipeline for request understanding, floor localization, object detection, and action-decision making. To support research and evaluation in this new setting, we construct the Vision-Language Delivery (VLD) dataset within the CARLA simulator. Experimental results on the VLD dataset showcase the feasibility of the LogisticsVLN system. In addition, we conduct subtask-level evaluations of each module of our system, offering valuable insights for improving the robustness and real-world deployment of foundation model-based vision-language delivery systems.

LogisticsVLN: Vision-Language Navigation For Low-Altitude Terminal Delivery Based on Agentic UAVs

TL;DR

LogisticsVLN tackles the problem of precise, window-level terminal delivery using autonomous UAVs by integrating lightweight multimodal language and vision models into a modular VLN pipeline. The approach grounds natural language requests into target floors and windows via a Floor Count VLM and a target descriptor through a Request Understanding module, then performs object grounding and safe exploration with a Depth Assistant and a Viewpoint Selection strategy. A dedicated Visual-Language Delivery (VLD) dataset in CARLA, plus ablation analyses, demonstrate the system’s feasibility and identify bottlenecks in floor localization and object recognition across different VLMs. The work advances practical aerial delivery by providing a deployable, training-free framework and a benchmark for evaluating VLM-driven aerial delivery tasks in unseen environments.

Abstract

The growing demand for intelligent logistics, particularly fine-grained terminal delivery, underscores the need for autonomous UAV (Unmanned Aerial Vehicle)-based delivery systems. However, most existing last-mile delivery studies rely on ground robots, while current UAV-based Vision-Language Navigation (VLN) tasks primarily focus on coarse-grained, long-range goals, making them unsuitable for precise terminal delivery. To bridge this gap, we propose LogisticsVLN, a scalable aerial delivery system built on multimodal large language models (MLLMs) for autonomous terminal delivery. LogisticsVLN integrates lightweight Large Language Models (LLMs) and Visual-Language Models (VLMs) in a modular pipeline for request understanding, floor localization, object detection, and action-decision making. To support research and evaluation in this new setting, we construct the Vision-Language Delivery (VLD) dataset within the CARLA simulator. Experimental results on the VLD dataset showcase the feasibility of the LogisticsVLN system. In addition, we conduct subtask-level evaluations of each module of our system, offering valuable insights for improving the robustness and real-world deployment of foundation model-based vision-language delivery systems.
Paper Structure (17 sections, 7 equations, 3 figures, 5 tables)

This paper contains 17 sections, 7 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Visual-Language Navigation
  4. MLLM-based UAV systems
  5. Autonomous delivery
  6. LogisticsVLN
  7. Task Definition
  8. System Overview
  9. Request Understanding
  10. Floor Localization
  11. Object Exploration
  12. Experiments
  13. VLD Dataset
  14. Evaluation Metrics and Implementation Details
  15. Results
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Overview of the LogisticsVLN system based on an agentic UAV
  • Figure 2: Statistics of VLD Dataset: (a) target categories, (b) building types, (c) floor number distribution, and (d) difficulty distribution.
  • Figure 3: Examples and case studies about the effect of Choice VLM: (a) and (b) are examples of ideal viewpoint selection, while (c) and (d) are two case studies of suboptimal viewpoint.