Multi-Objective Quality-Diversity in Unstructured and Unbounded Spaces
Hannah Janmohamed, Antoine Cully · 2025
Read Multi-Objective Quality-Diversity in Unstructured and Unbounded Spaces by Hannah Janmohamed, Antoine Cully (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-03715.txt.
Introduction
Hannah Janmohamed and Antoine Cully wrote this paper in two thousand twenty-five. It extends quality-diversity search, which hunts for many kinds of good answers, to spaces with no grid and no bounds.
Multi-goal quality-diversity weighs trade-offs, such as speed against energy use, while keeping variety. Grid methods must tile the trait space first, which fails when traits are learned or unbounded. The new method, MOUR-QD, keeps its stock without a grid. On five robot tasks it matches set rivals and at times doubles their score. It shines where traits must be learned on the fly, a door to protein design and image work.
Catalog record
| CREATOR | Hannah Janmohamed, Antoine Cully |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-03715
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Multi-Objective Quality-Diversity in Unstructured and Unbounded Spaces Hannah Janmohamed, Antoine Cully Abstract . Quality-Diversity algorithms are powerful tools for discovering diverse, high-performing solutions. Recently, Multi-Objective Quality-Diversity ( moqd ) extends qd to problems with several objectives while preserving solution diversity. moqd has shown promise in fields such as robotics and materials science, where finding trade-offs between competing objectives like energy efficiency and speed, or material properties is essential. However, existing methods in moqd rely on tessellating the feature space into a grid structure, which prevents their application in domains where feature spaces are unknown or must be learned, such as complex biological systems or latent exploration tasks. In this work, we introduce Multi-Objective Unstructured Repertoire for Quality-Diversity ( mour-qd ), a moqd algorithm designed for unstructured and unbounded feature spaces. We evaluate mour-qd on five robotic tasks. Importantly, we show that our method excels in tasks where features must be learned, paving the way for applying moqd to unsupervised domains. We also demonstrate that mour-qd is advantageous in domains with unbounded feature spaces, outperforming existing grid-based methods. Finally, we demonstrate that mour-qd is competitive with established moqd methods on existing moqd tasks and achieves double the moqd-score in some environments. mour-qd opens up new opportunities for moqd in domains like protein design and image generation. DOI: 10.1145/3712256.3726394 ISBN: 979-8-4007-1465-8/2025/07 Conference: The Genetic and Evolutionary Computation Conference 2025; July 14–18, 2025; Málaga, Spain CCS: Theory of computation Evolutionary algorithms CCS: Applied computing Multi-criterion optimization and decision-making CCS: Computer systems organization Evolutionary robotics 1. Introduction Figure 1 . Left: Existing moqd methods store Pareto Fronts in each cell of a MAP-Elites grid. Right: mour-qd introduces an unstructured archive, storing solutions in a continuous manner throughout the feature space. Pareto Fronts are defined locally, using a radius 2l around each solution. In recent years, Quality-Diversity ( qd ) algorithms ( Mouret and Clune, 2015 ; Cully and Demiris, 2018 ) have gained significant attention for their ability to discover a diverse set of high-performing solutions. Unlike traditional optimisation methods that focus on identifying a single optimal solution, qd methods aim to uncover a collection of solutions that vary across different behaviours or features while still performing well. This ability to balance exploration and optimisation has proven valuable across a range of real-world, complex applications. For example, in robotics, qd has been applied to damage recovery by providing fallback behaviours that allow robots to adapt to hardware failures, such as broken limbs or motors ( Cully et al., 2015 ; Allard et al., 2022 ) . Alternatively, qd has been used to generate diverse adversarial prompts in order to train Large Language Models, helping identify vulnerabilities and improve model safety by testing the system across a broad range of inputs ( Samvelyan et al., 2024 ) . Building on the foundations of qd , Multi-Objective Quality-Diversity ( moqd ) extends this framework to tackle problems where solutions must balance multiple competing objectives while maintaining behavioural diversity ( Pierrot et al., 2022 ; Janmohamed et al., 2023 ; Janmohamed et al., 2024a ) . This approach has proven valuable across a range of applications too. In robotics, moqd enables exploration of trade-offs between objectives such as energy consumption and speed, or fitness and reproducibility, while preserving a diverse repertoire of behaviours ( Pierrot et al., 2022 ; Janmohamed et al., 2023 ; Janmohamed et al., 2024a ; Flageat et al., 2024 ) . In materials design, it facilitates the discovery of materials which achieve different trade-offs across material objectives, such as strength, durability, and conductivity, with diverse properties, providing alternative options if some materials prove impractical to synthesise ( Janmohamed et al., 2024b ) . In architecture, moqd supports the generation of building designs that balance objectives like ventilation efficiency and noise reduction, while offering multiple diverse options to suit different requirements or preferences ( Gaier et al., 2022 ) . moqd plays a critical role in solving problems that require optimisation across multiple objectives while preserving diversity, making it a powerful tool for tackling complex and multi-faceted challenges. Despite these advances, existing moqd methods rely on discretising the feature space into a structured map-elites grid, where each cell contains a Pareto front as illustrated in Figure 1 . While effective, this approach assumes that the features of the tasks can be hand-defined and that their bounds are known beforehand. These assumptions can be restrictive, particularly in scenarios where the defining features of solutions are unclear or difficult to specify, or when the bounds of the feature space are unknown. For example, tasks requiring latent exploration ( Cully, 2019 ; Grillotti and Cully, 2022b ; Fontaine and Nikolaidis, 2021 ) are often unsuitable for grid-based qd algorithms because the shape of the latent space is unknown in advance and could dynamically change. In single-objective qd , unstructured archives have been used to enable automatic feature learning ( Cully, 2019 ; Grillotti and Cully, 2022b ) , allowing solutions to be organised without predefined feature boundaries. However, adapting this concept to the multi-objective setting is particularly challenging. In moqd , the grid structure provides a natural way to organise and maintain Pareto fronts within each cell. Without the grid, it becomes unclear how to define Pareto fronts in a continuous, unstructured feature space, posing a significant barrier to extending unstructured containers to multi-objective problems. As a result, no existing methods address these challenges in moqd , which limits their applicability to problems requiring latent space exploration or unsupervised feature discovery. This paper presents Multi-Objective Unstructured Repertoire for Quality-Diversity ( mour-qd ), a novel moqd algorithm that uses a multi-objective unstructured archive, as illustrated in Figure 1 . Unlike existing moqd approaches, mour-qd does not rely on predefined features or fixed feature space bounds. Instead, it stores solutions in an unstructured and unbounded manner and defines Pareto Fronts locally around solutions. We evaluate mour-qd across 5 continuous control robotics tasks and show that it offers three key advantages. First, we evaluate mour-qd on existing moqd tasks and demonstrate that it is competitive with current moqd baselines and doubles the moqd-score of current moqd baselines in some environments. This confirms that mour-qd is an effective solution for advancing existing moqd applications. Second, we show that mour-qd improves on current moqd baselines in its robustness to cases where feature space bounds are unknown. By removing the need for prior knowledge of feature space limits, mour-qd makes moqd algorithms more flexible and adaptable to a wider range of complex, real-world problems. Finally, we show that mour-qd enables moqd to operate in unsupervised settings by learning features directly from data. Crucially, this capability opens the door for applying moqd to entirely new domains, such as protein design and latent space exploration, where manually defining features is impractical. By addressing these challenges, mour-qd represents a meaningful step forward in the evolution of moqd , unlocking new opportunities for exploring diversity and trade-offs in challenging optimisation problems. All of our code was implemented using the QDax codebase ( Lim et al., 2022 ; Chalumeau et al., 2023 ) which uses massive parallelisation via Jax ( Bradbury et al., 2018 ) . Our code was fully containerised and is publicly available at https://github.com/adaptive-intelligent-robotics/MOUR-QD .