loading page

SwarmL: UAV swarm task description language with AI policies enhancement
  • +2
  • Shaocong Ma,
  • Liang Wang,
  • Wen Wang,
  • Hao Hu,
  • Xianping Tao
Shaocong Ma
State Key Laboratory of Novel Software Technology
Author Profile
Liang Wang
State Key Laboratory of Novel Software Technology

Corresponding Author:wl@nju.edu.cn

Author Profile
Wen Wang
State Key Laboratory of Novel Software Technology
Author Profile
Hao Hu
State Key Laboratory of Novel Software Technology
Author Profile
Xianping Tao
State Key Laboratory of Novel Software Technology
Author Profile

Abstract

In recent years, unmanned aerial vehicle (UAV) swarms have gained traction in diverse fields, such as disaster relief and agricultural production. However, programming tasks for UAV swarms presents significant challenges due to their scale, task complexity, and the need for autonomy and collaboration. To address these issues, we propose SwarmL, a task description language specifically designed for orchestrating UAV swarm operations, accompanied by its runtime interpreter. SwarmL features several key attributes: (1) a clear task decomposition framework with a “Main-Task-Behavior-Action” structure and “init-goal-routine” control flow; (2) integration of trained AI policies into swarm task execution; (3) support for concurrent operations through language primitives like “each” and “||”, enabling concise descriptions of multi-level concurrency among groups, UAVs, and devices; and (4) collaborative capabilities via a blackboard system that facilitates autonomous communication among UAVs. To illustrate the usability of SwarmL and its interpreter, we present a search-and-delivery task within the Airsim simulator, successfully integrating policies from the DQN reinforcement learning algorithm. This example demonstrates effective concurrent management, allowing multiple UAVs to execute tasks simultaneously while enhancing cooperation between search and delivery UAVs through the blackboard system.
16 Nov 2024Submitted to Software: Practice and Experience
17 Nov 2024Submission Checks Completed
17 Nov 2024Assigned to Editor
18 Nov 2024Review(s) Completed, Editorial Evaluation Pending
06 Dec 2024Reviewer(s) Assigned