QED: using Quality-Environment-Diversity to evolve resilient robot swarms

TL;DR

This work tackles fault resilience in swarm robotics with a model-free approach. It introduces Quality-Environment-Diversity (QED), a MAP-Elites–based framework that evolves swarm controllers across diverse, randomly perturbed environments and indexes solutions by environment type rather than traditional behavioural descriptors. Through comparisons with a hand-coded descriptor, SDBC, and SPIRIT across five swarm tasks, QED demonstrates improved fault-recovery resilience and generates a diverse repertoire of high-performing recovery solutions, with fault impact typically reduced by a factor of 2–3. The study further shows that environment-informed diversity enhances robustness to high-impact faults, though it may trade off performance under normal operating conditions in some tasks. Overall, QED offers a scalable, domain-agnostic approach to fault recovery in swarm robotics and highlights the value of explicit environmental diversity in quality-diversity optimization.

Abstract

In swarm robotics, any of the robots in a swarm may be affected by different faults, resulting in significant performance declines. To allow fault recovery from randomly injected faults to different robots in a swarm, a model-free approach may be preferable due to the accumulation of faults in models and the difficulty to predict the behaviour of neighbouring robots. One model-free approach to fault recovery involves two phases: during simulation, a quality-diversity algorithm evolves a behaviourally diverse archive of controllers; during the target application, a search for the best controller is initiated after fault injection. In quality-diversity algorithms, the choice of the behavioural descriptor is a key design choice that determines the quality of the evolved archives, and therefore the fault recovery performance. Although the environment is an important determinant of behaviour, the impact of environmental diversity is often ignored in the choice of a suitable behavioural descriptor. This study compares different behavioural descriptors, including two generic descriptors that work on a wide range of tasks, one hand-coded descriptor which fits the domain of interest, and one novel type of descriptor based on environmental diversity, which we call Quality-Environment-Diversity (QED). Results demonstrate that the above-mentioned model-free approach to fault recovery is feasible in the context of swarm robotics, reducing the fault impact by a factor 2-3. Further, the environmental diversity obtained with QED yields a unique behavioural diversity profile that allows it to recover from high-impact faults.

Figures (3)

• Figure 1: Thymio II robot augmented with a Raspberry Pi 3 B+ model.
• Figure 2: Map quality analysis based on behavioural diversity and performance: (a) development of the map coverage (Mean $\pm$ SD) over generations, with solid line indicating the average across all 25 independently evolved maps (5 independent replicates for each of 5 tasks), and shaded region indicating the standard-deviation across replicates (averaged across all 5 tasks). (b) boxplot of the best performance, evaluated in the normal operating environment; to obtain the best performance data for a given map, all its individual solutions are re-evaluated for 10 independent trials in the normal operating environment; for any given algorithm, each data point in the box plot represents one of the 25 re-evaluated maps.
• Figure 3: Visualisation of the fault recovery data based on impact of the fault, resilience, and behavioural diversity around the normal behaviour: (a) impact-resilience signature, illustrating the probability distribution of resilience as a function of the impact of the fault, with the dashed line indicating the identity line. An ideal signature has most of its probability mass in the top right corner, where faults do not affect the swarm's performance; however, in case there are high-impact faults, ideally the resilience should be constant across the impact spectrum, indicating that recovery will be possible even for high-impact faults. (b) diversity-resilience signature, illustrating the probability distribution of behavioural diversity, based on the behavioural distance to the normal behaviour, as a function of resilience. An ideal signature has high resilience and covers the entire distance spectrum, indicating a strong fault recovery profile with useful diversity. (c) diversity-impact signature, illustrating the probability distribution of behavioural diversity, based on the behavioural distance to the normal behaviour, as a function of the impact of the fault. Each of the signatures illustrates a correlation between the impact of the fault and the behavioural diversity, indicating behaviours that differ strongly from the normal behaviour allow improved recovery from high-impact faults.