Population Dynamics in ARIEL Robotics Systems Featuring Embodied Evolution via Spatial Mating Mechanisms Conference: Genetic and Evolutionary Computation Conference; July 2026; San José, Costa Rica Proceedings of the Genetic and Evolutionary Computation Conference (GECCO ’26), July 2026, San José, Costa Rica CCS: Computing methodologies Evolutionary robotics

Victoria Peterson, Akshat Srivastava, Raghav Prabhakar · 2026

Read Population Dynamics in ARIEL Robotics Systems Featuring Embodied Evolution via Spatial Mating Mechanisms Conference: Genetic and Evolutionary Computation Conference; July 2026; San José, Costa Rica Proceedings of the Genetic and Evolutionary Computation Conference (GECCO ’26), July 2026, San José, Costa Rica CCS: Computing methodologies Evolutionary robotics by Victoria Peterson, Akshat Srivastava, Raghav Prabhakar (2026) 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-2604-26822.txt.

Introduction

Peterson, Srivastava and Prabhakar test how physical layout changes evolution in a mock robot group. Their gecko-like four-legged ARIEL robots, driven by evolved neural nets, must travel to find mates and survive under rules based on place. Mating with near neighbors barely changed peak fitness, while pairing place-based parent choice with random deaths made the group unstable. Deaths tied to density finished most runs but lowered fitness, showing that only fitness-based choice kept the group stable.

Catalog record

CREATORVictoria Peterson, Akshat Srivastava, Raghav Prabhakar
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37657
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.26822

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Population Dynamics in ARIEL Robotics Systems Featuring Embodied Evolution via Spatial Mating Mechanisms Conference: Genetic and Evolutionary Computation Conference; July 2026; San José, Costa Rica Proceedings of the Genetic and Evolutionary Computation Conference (GECCO ’26), July 2026, San José, Costa Rica CCS: Computing methodologies Evolutionary robotics Victoria Peterson, Akshat Srivastava, Raghav Prabhakar Abstract . We present a Spatially Embedded Evolutionary Algorithm where robot individuals exist in a physically simulated 2D environment, must navigate to encounter potential mates, and compete for survival under various spatially-aware selection pressures. Using HyperNEAT evolved neural controllers for ARIEL gecko-inspired quadrupeds in MuJoCo, we investigate how spatial structure fundamentally alters evolutionary dynamics. Our experiments show a modest 4.9% difference in peak fitness between proximity-based and random pairing—possibly within stochastic variation—while combining spatial parent selection with stochastic death selection produces unstable population dynamics. We discover a continuous phase transition in energy-based selection experiments, with critical zone count n c ≈ 14.7 n_{c}\approx 14.7 separating extinction-dominated and explosion-dominated regimes. Our density-dependent death selection mechanism achieves 97% completion rates but causes fitness decline, revealing a fundamental dilemma where decoupled mechanisms produce bistable dynamics, positively coupled mechanisms create counter-selection pressures, and only deterministic fitness-based selection maintains stability. These findings provide important constraints for future spatial EA design. 1. Introduction Evolutionary Algorithms (EAs) have been widely used to evolve robotic controllers and morphologies, typically under abstract evolutionary assumptions such as panmictic populations, where any individual may reproduce with any other. While computationally convenient, these abstractions ignore spatial constraints that are fundamental to biological evolution, where interactions are local and shaped by physical proximity. Such constraints introduce localized selection pressure, genetic drift, and population clustering, all of which influence evolutionary dynamics and emergent behavior. In contrast, spatially structured evolutionary systems embed individuals in space and restrict interactions to local neighborhoods. This induces isolation by distance , slowing the global diffusion of high-fitness genotypes and promoting the emergence of spatial niches. Prior work has shown that these dynamics can preserve diversity and mitigate premature convergence, but also lead to heterogeneous fitness distributions and localized behavioral specialization compared to globally mixing populations ( Chopard et al. ; pap ) . This work investigates spatial evolution in the context of embodied robotics, where evolutionary processes unfold through the physical interactions of autonomous agents rather than centralized evaluation. We present a Spatial Evolutionary Algorithm in which ARIEL robot individuals exist in a physically simulated 2D environment, must navigate to encounter potential mates, and compete for reproduction under spatially mediated selection pressures. We tightly integrate evolutionary computation with physics-based simulation via ARIEL MuJoCo, where spatial dynamics, movement, and resource constraints jointly shape adaptation. In this setting, the environment itself implicitly applies selection pressure, shifting the evolutionary process from objective-driven optimization toward open-ended ecological interaction ( don ; Bredeche et al. b ; Bredeche et al. a ) . Research Questions Our investigation addresses four key questions at the intersection of spatial evolution and embodied robotics: RQ1: How does spatial structure affect Evolutionary Algorithm Dynamics? RQ2: How do different selection pressures impact adaptation in Spatial Environments? RQ3: How can non-spatial Parent Selection Operators be modified to apply spatially? RQ4: Can spatial Parent Selection and Death Selection operators be designed to balance population growth? 2. Related Work This research builds on three intersecting lines of work: spatially structured evolutionary algorithms, embodied evolution, and population dynamics in decentralized evolutionary systems. 2.1. Spatially Structured Evolutionary Algorithms To mitigate premature convergence in globally mixing populations, spatially structured EAs keep interactions in local sub-populations, suing approaches such as coarse-grained Island Models, where distinct populations exchange individuals via migration, and fine-grained Cellular EAs, where individuals interact only with immediate neighbors on a grid ( Sudholt ; Skolicki ; kro ) . Spatial restriction slows the spread of advantageous genotypes, maintaining diversity through localized competition and reduced gene flow. Our work extends these concepts beyond fixed topologies to continuous physical space, where neighborhoods emerge dynamically in flat 2D space through movement and proximity rather than predefined grid or island structures. 2.2. Embodied Evolution and Interactive Ecosystems Embodied Evolution (EE) distributes evolutionary processes across populations of autonomous robots, eliminating centralized evaluation and conducting selection and reproduction online during system operation ( don ; Bredeche et al. a ) . This paradigm has evolved into Interactive Robot Ecosystems, such as mEDEA, where agents continuously exchange genetic material through local encounters rather than explicit fitness ranking ( Russo et al. 2022 ) . In these systems, survival and reproductive success depend on an agent’s ability to remain operational within its environment, so selection pressure is implicit and environment-driven, with outcomes affected by navigation ability, energy management, and encounter frequency. Rather than optimizing a predefined objective, adaptation emerges from sustained interaction between agents and their surroundings, making EE good for studying spatial evolutionary dynamics. 2.3. Mating and Population Dynamics Mate selection in decentralized evolutionary systems is opportunistic and local, relying on physical encounters rather than global fitness comparisons, which helps preserve diversity, but can also favor behaviors that maximize encounter rates or clustering over exploration ( Diependaal ) . Unlike traditional EAs with fixed population sizes, embodied ecosystems exhibit dynamic populations, while energy is often used as a proxy for fitness, where agents must balance survival costs with reproductive investment. This coupling of energy, movement, and reproduction creates a natural mechanism for population regulation while introducing trade-offs between exploration, mating success, and longevity ( Yao et al. ; Bredeche et al. b ) . 3. Theory This section outlines the theoretical basis and core mechanisms of our spatial evolutionary system: spatially structured EAs, HyperNEAT control representations, and the parent/death selection operators that govern reproduction and survival in a 2D embodied environment. Table 1 summarizes the notation used throughout. Table 1. Summary of notation used in this work. Symbol Description Section HyperNEAT / Controller w,w_{ij} Connection weight (from neuron i to j ) 3.2 (x_{1},y_{1},x_{2},y_{2}) Source and target neuron coordinates 3.2 n Number of robot joints 3.2 \tau Weight pruning threshold 3.2 Spatial Structure \mathbf{x}_{i} Position of individual i in world space 3.3 d_{ij} Periodic distance between individuals i and j 3.3 W,H World width and height 3.6 Parent Selection r_{\text{pair}} Pairing radius for proximity mating 3.3 r_{\text{zone}} Radius of mating zones 3.3 \mathbf{c}_{k} Center position of mating zone k 3.3 P Set of already-paired individuals 3.3 n_{\text{gen}} Generations between zone relocations 3.3 \mathbf{d}_{\text{input}} Directional input vector to controller 3.4 Death Selection N Target population size 3.5 \mathrm{age}_{i},\mathrm{age}_{\max} Age of individual i ; maximum age 3.5 E_{i} Energy level of individual i 3.5 \delta E_{\text{depletion}} Energy lost per generation 3.5 \Delta E_{\text{mating}} Energy change from mating 3.5 \rho_{i} Local density around individual i 3.5 \rho_{c} Critical density threshold 3.5 \sigma Locality radius (density kernel width) 3.5 P_{\text{base}},P_{\text{max}} Base and max death probabilities 3.5 Fitness Evaluation f_{\text{simple}},f_{\text{directional}} Fitness functions 3.7 \mathbf{x}_{\text{start}},\mathbf{x}_{\text{end}} Start and end positions 3.7 d_{\text{total}} Total distance traveled 3.7 w_{\text{progress}} Weight for directional bonus 3.7 b_{\text{direction}} Directional progress score 3.7 \mathbf{u}_{\text{target}} Unit vector toward target 3.7 3.1. Evolutionary Algorithms and Spatial Structure Conventional EAs assume panmictic populations where any individual may mate with any other, where-as spatial EAs embed individuals in space and restrict interaction to local neighborhoods. These systems exhibit reduced global gene flow and spatial heterogeneity, with

Read the complete text