AT-Drone: Benchmarking Adaptive Teaming in Multi-Drone Pursuit

TL;DR

AT-Drone addresses the challenge of enabling drones to collaborate with unseen teammates in real-world pursuit tasks. It formalizes the problem as an extended AT-Dec-POMDP defined by $({\mathcal{S}}, {\mathcal{C}}, {\mathcal{A}}, {\mathcal{P}}, r, {\mathcal{O}}, \gamma, T)$ and provides a four-part benchmark: customizable simulations, real-world deployment pipelines, a distributed algorithm zoo including HOLA-Drone V2 and NAHT-D, and standardized unseen-teammate evaluations. Four progressively challenging environments and three unseen drone zoos enable robust benchmarking across diverse coordination tasks, with both zero-shot and ad-hoc teamwork settings evaluated. Real-world Crazyflie experiments validate practicality and highlight significant gains from advanced adaptive-teaming methods, suggesting strong potential to improve reliability and efficiency of real-world drone operations.

Abstract

Adaptive teaming-the capability of agents to effectively collaborate with unfamiliar teammates without prior coordination-is widely explored in virtual video games but overlooked in real-world multi-robot contexts. Yet, such adaptive collaboration is crucial for real-world applications, including border surveillance, search-and-rescue, and counter-terrorism operations. To address this gap, we introduce AT-Drone, the first dedicated benchmark explicitly designed to facilitate comprehensive training and evaluation of adaptive teaming strategies in multi-drone pursuit scenarios. AT-Drone makes the following key contributions: (1) An adaptable simulation environment configurator that enables intuitive and rapid setup of adaptive teaming multi-drone pursuit tasks, including four predefined pursuit environments. (2) A streamlined real-world deployment pipeline that seamlessly translates simulation insights into practical drone evaluations using edge devices and Crazyflie drones. (3) A novel algorithm zoo integrated with a distributed training framework, featuring diverse algorithms explicitly tailored, for the first time, to multi-pursuer and multi-evader settings. (4) Standardized evaluation protocols with newly designed unseen drone zoos, explicitly designed to rigorously assess the performance of adaptive teaming. Comprehensive experimental evaluations across four progressively challenging multi-drone pursuit scenarios confirm AT-Drone's effectiveness in advancing adaptive teaming research. Real-world drone experiments further validate its practical feasibility and utility for realistic robotic operations. Videos, code and weights are available at \url{https://sites.google.com/view/at-drone}.

Figures (8)

• Figure 1: Overview of the AT-Drone Benchmark, comprising four key components: (I) a customizable simulation environment featuring varied multi-drone pursuit tasks with adjustable complexity; (II) a streamlined real-world deployment pipeline employing motion capture systems and edge devices to facilitate realistic drone validation; (III) a distributed training framework equipped with diverse adaptive teaming algorithms for multi-drone pursuit task; and (IV) standardized evaluation protocols, leveraging diverse unseen teammate configurations to rigorously evaluate adaptive teaming performance and robustness across distinct strategies.
• Figure 2: Success rate (SUC) across different difficulty levels for adaptive teaming without teammate modelling. Red dotted lines denote best-response baselines specifically trained on the given unseen teammate zoo.
• Figure 3: Performance comparison of adaptive teaming with teammate modeling across environments with varying difficulties.
• Figure 4: Illustration of four multi-drone pursuit environments in real world. The environments vary in the number of pursuers (p), evaders (e), and obstacles (o), denoted as 4p2e3o, 4p2e1o, 4p2e5o, and 4p3e5o. Each setup introduces different levels of complexity, testing the adaptability and coordination capabilities of the agents.
• Figure 5: An example of environment configuration file.

Theorems & Definitions (2)

• Definition 3.1: Adaptive Teaming in Multi-Drone Pursuit
• Definition C.1: Preference Optimal