Improved Swarm Engineering: Aligning Intuition and Analysis

TL;DR

This paper introduces new metrics to assist in designing and analyzing swarm-robotic systems, improving prediction accuracy of swarm behavior and reducing design costs through quantitative analysis of control algorithms.

Contribution

It develops novel metrics for swarm self-organization, scalability, flexibility, and robustness, validated through analysis of multiple control algorithms in different scenarios.

Findings

Metrics accurately predict algorithm performance in different scenarios.

Quantitative results support intuitive assessments of swarm behaviors.

Metrics can be combined to improve behavior prediction.

Abstract

We present a set of metrics intended to supplement designer intuitions when designing swarm-robotic systems, increase accuracy in extrapolating swarm behavior from algorithmic descriptions and small test experiments, and lead to faster and less costly design cycles. We build on previous works studying self-organizing behaviors in autonomous systems to derive a metric for swarm emergent self-organization. We utilize techniques from high performance computing, time series analysis, and queueing theory to derive metrics for swarm scalability, flexibility to changing external environments, and robustness to internal system stimuli such as sensor and actuator noise and robot failures. We demonstrate the utility of our metrics by analyzing four different control algorithms in two scenarios: an indoor warehouse object transport scenario with static objects and a spatially unconstrained outdoor search and rescue scenario with moving objects. In the spatially constrained warehouse scenario, efficient use of space is key to success so algorithms that use mechanisms for traffic regulation and congestion reduction are the most appropriate. In the search and rescue scenario, the same will happen with algorithms which can cope well with object motion through dynamic task allocation and randomized search trajectories. We show that our intuitions about comparative algorithm performance are well supported by the quantitative results obtained using our metrics, and that our metrics can be synergistically used together to predict collective behaviors based on previous results in some cases.