Trade-offs of Dynamic Control Structure in Human-swarm Systems

TL;DR

This paper tackles the trade-offs between centralised and decentralised control in human-swarm systems for environmental monitoring. It proposes a simple, dynamic hybrid coordination framework that forms a forest of trees controlled by four mechanisms—Root Formation, Tree Growth, Active Recruitment, and Dissolution—and uses a threshold $\delta$ to balance centralisation versus decentralisation. Empirical results show the hybrid approach can outperform purely centralised strategies in task performance by about $19.2\%$ and reduce operator communications by about $23.1\%$, with ablation studies underscoring the importance of Active Recruitment for maintaining swarm adaptivity. The work suggests that adaptive centralisation, potentially via learning to adjust $\delta$, can yield robust performance and lower human workload in real-world multi-robot systems, while also highlighting the need to manage inter-swarm communication overhead.

Abstract

Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.

Figures (5)

• Figure 1: Examples the decentralised approach and early tree formations of the hierarchical and hybrid approaches during the task. The different coloured agents represent members of a tree at different depths, with root nodes being coloured blue. The small black circles represent existing incomplete events and the underlying event density distribution is shown.
• Figure 2: Box plot of waiting times for environmental events to be observed by a swarm of 25 agents coordinated with the Hierarchical, Decentralised and Hybrid approaches. Event waiting time data was averaged across each replicate, and aggregated across 20 replicates.
• Figure 3: Mean number of agents that are part of trees by swarms of sizes 25 and 100 agents coordinated with the hybrid approach with $\delta = 3$. Data was aggregated across 20 replicates.
• Figure 4: Ablation analysis of mean waiting times for events to be observed by a swarm of 25 hierarchically coordinated agents following the removal of the active recruitment (AR), and the dissolution elements.
• Figure 5: Mean number of messages sent by different agents in a swarm of size 25 to a human operator against the mean number of messages sent at each timestep within the swarm.