This paper presents a new approach for generating energy-efficient underwater glider motion parameters. First, a valid dataset is built by extracting key input features—path segments, ocean currents, and seafloor topography—while using pitch angle and diving depth as output variables. We then propose a neural architecture search-based back propagation neural network (NAS-BPNN) to enhance the model’s learning and generalization capabilities. In this approach, the number of hidden layers and neurons, activation functions, and training algorithms are treated as decision variables. Encoding-decoding strategies and a fitness evaluation method are designed for these variables. The success history-based adaptive differential evolution algorithm is applied to optimize the network architecture. Simulation experiments demonstrate the effectiveness of the dataset and verify the importance of neural architecture search. Furthermore, NAS-BPNN outperforms the two latest BPNN methods, achieving the highest determination coefficients for motion parameters.

In nature, the emergence of highly intelligent and flexible behaviors from large groups of simple individuals is a common phenomenon. To uncover the hidden self-organizing rules behind these collective behaviors, existing studies have modeled various self-organizing systems to replicate different swarm motion patterns, often based on biological or physical observations and specific assumptions. However, few studies have adopted a data-driven approach to model collective self-organization rules. In this paper, we leverage domain knowledge to propose a data-driven method to learn self-organization rules directly from collective motion data. We first construct a foundational collective model template that integrates existing domain knowledge and can represent various collective behaviors through different parameter configurations. Building on this template, we developed a bilevel nested evolutionary framework based on a genetic algorithm (GA), named BLGA, to optimize parameter combinations within the template. Using the template, we designed four distinct collective models for particle simulations, successfully replicating behaviors such as swarming and flocking, and collected corresponding motion data. The BLGA framework was then applied to learn from these data. Through interpreting evolutionary models, reproducing collective behaviors, and conducting a comparative analysis of evaluation metrics, our results demonstrate that the proposed method effectively captures the hidden self-organizing rules underlying the collective motion data.

This article presents a distributed control strategy for herding groups of evaders towards a predefined goal region using a team of robotic herders. In herding problems, evaders tend to move away from each other to increase their coverage regions. This makes it challenging to develop control solutions since the wandering evaders need to be collected while driving the herd. To address this, we propose the distributed outmost push strategy, where each robotic herder pushes the evader that is farthest from the goal region. The intuition behind this strategy is that robotic herders should focus on evaders that are further from the goal region as they are more likely to be missed during the herding process. The outmost evaders are selected from the local field of view, and the robotic herders make decisions in a decentralized manner. We also analyze the convergence of the designed dynamics and the minimum sensing range required for herders. The proposal's effectiveness and generality are validated through numerical simulations and real robotic experiments.

Homogeneous swarm robots are of significant research interest due to their robustness, flexibility, and scalability in completing complex tasks across various applications. This paper focuses on trapping heterogeneous targets using swarm robots, with emphasis on their different strengths. These targets consist of weak, strong, and group-moving individuals, where stronger targets exhibit larger body size, higher physical strength, and stronger resistance ability. Our goal is to develop an adaptive controller that enables swarm robots to self-organize and distribute themselves to trap targets, while adjusting encirclement thickness and robot group size based on target strength. We leverage local implicit information generated from density-based interaction to improve inter-robot and robot-target interactions through adaptive allocation and transformation mechanisms. The feasibility of our approach is validated through numerical simulations and experiments involving up to 50 physical robots and one human-controlled transformer.

Many real-world optimization tasks suffer from noise. So far, the research on noise-tolerant optimization algorithms is still restricted to low-dimensional problems with less than 100 decision variables. In reality, many problems are high dimensional. Cooperative coevolutionary (CC) algorithms based on a divide-and-conquer strategy are promising in solving complex high-dimensional problems. However, noisy fitness evaluations pose a challenge in problem decomposition for CC. The state-of-the-art grouping methods, such as differential grouping (DG) and recursive DG, are unable to work properly in noisy environments. Because it is impossible to distinguish whether the change of one variable’s difference value is caused by noise or the perturbation of its interacting variables. As a result, every pair of variables will be identified as nonseparable in these methods. In this article, we study how to group decision variables with the covariance matrix adaptation evolution strategy (CMA-ES) in noisy environments and subsequently propose a landscape-aware grouping (LAG) method. Instead of detecting pairwise interacting variables, we directly identify a nonseparable subcomponent. To this end, we propose to use two convergence features: 1) variable convergence time and 2) accumulative path, to describe variables’ fitness landscapes; then, variables are clustered according to these two features. Numerical experiments show that LAG can more effectively identify interactive decision variables in the presence of multiplicative noise than the DG and some of its variants. Up to 500 dimensions, the performance of CC CMA-ES with landscape-aware grouping (CC-CMAES-LAG) is competitive compared with existing CC algorithms and uncertainty-handling CMA-ES (UH-CMA-ES).

A swarm of robots is a typical self-organizing system. Parameters of self-organizing model can regulate the interaction among the individuals to achieve the desired collective motion. Thus, it makes sense to optimize the parameters for better performance in self-organization. The evolutionary algorithm (EA) is one promising heuristic optimization algorithm. However, quantitative evaluation of a self-organizing model is still an open scientific question. In this paper, we proposed three metrics for collective motion from the perspective of grouping, polarization and spatial distribution of group. Then, we designed the fitness function with the proposed metrics to bridge the gap between EA and parameter tuning of self-organizing model. In the experimental study, we optimized two kinds of self-organizing models using the differential evolution algorithm (DE) and validated the effectiveness in simulation. Moreover, we validated the optimized parameters on a swarm of physical robots. The experimental results show that the proposed metrics are effective to quantitatively evaluate the self-organizing model so that the DE performs well for the automatic parameter tuning of controlling swarm robots.

Mobile agent path planning (MAPP) problem is a typical optimization problem. When we consider multiple agents path planning simultaneously, problems can be seen as multi-task optimization (MTO) problems. The Multi-factorial evolutionary algorithm (MFEA) is one promising technique for MTO problems. Within the MFEA some selective individuals that contain useful knowledge are transferred among independent tasks to enhance the convergence. In this work, we investigate what information, except to the selective individuals, should be transferred under the framework of the MFEA. In particular, the difference between the individuals and the estimated optimal solution of the corresponding task is used to calculate individual_gradient (IG), which is introduced into the MFEA as additional knowledge for transferring. Empirical studies on nine benchmarks validate the effectiveness of IG based MFEA (MFEA-IG). Moreover, we apply the MFEA-IG to MAPP problems. Simulation results show that the MFEA-IG outperforms the original MFEA and single task EA.

There are many swarms of creatures in nature, which lead to a lot of highly ordered and beautiful "emergence" behaviors. For the modeling of self-organizing rules, most of the existing literature focuses on modeling according to special knowledge of physics or biology. Some self-organizing models are proposed in the literature which has been validated by the reproduction of certain "emergence" motion pattern, such as torus, or flocking. However, there are few studies about datadriven modeling of self-organizing rules of swarms. In this paper, we propose a prior knowledge free (i.e., data-driven) approach to learn the self-organizing rules of moving swarms. We use a Genetic Programming (GP) based two-layer framework to optimize the self-organizing model which is consist of neighbor selection rules and corresponding reaction rules. The proposed data-driven modeling method is validated by modeling of three typical collective behaviors (highly parallel group, dynamic parallel group and torus swarm behavior) only according to the simulation data generated from Vicsek and Couzin models. An analysis is conducted with expression tree simplification, swarm behavior reproduction and global metric evaluation. Results show that the proposed method can learn classic self-organizing rules effectively.

Cooperative coevolutionary (CC) algorithms decompose a problem into several subcomponents and optimize them separately. Such a divide-and-conquer strategy makes CC algorithms potentially well suited for large-scale optimization. However, decomposition may be inaccurate, resulting in a wrong division of the interacting decision variables into different subcomponents and thereby a loss of important information about the topology of the overall fitness landscape. In this paper, we suggest an idea that concurrently searches for multiple optima and uses them as informative representatives to be exchanged among subcomponents for compensation. To this end, we incorporate a multimodal optimization procedure into each subcomponent, which is adaptively triggered by the status of subcomponent optimizers. In addition, a nondominance-based selection scheme is proposed to adaptively select one complete solution for evaluation from the ones that are constructed by combining informative representatives from each subcomponent with a given solution. The performance of the proposed algorithm has been demonstrated by comparing five popular CC algorithms on a set of selected problems that are recognized to be hard for traditional CC algorithms. The superior performance of the proposed algorithm is further confirmed by a comprehensive study that compares 17 state-of-the-art CC algorithms and other metaheuristic algorithms on 20 1000-dimensional benchmark functions.

Cooperative co-evolution (CC) algorithm is a promising method to deal with large scale optimization (LSGO) problem. One major challenge in CC is to design a good decomposition strategy to decompose original problem into subproblems. Linkage learning is a technique to search the linkage betweens decision variables. In this paper, we proposed a dynamic online grouping CC algorithm which employ the linkage learning method of extended compact genetic algorithm (ECGA). This linkage learning method build marginal product models of population and search for models according to minimum description length (MDL) criterion. A discretization technique and normalization method of MDL value is used to make the linkage learning more adaptive in large scale optimization. The decision variables will be regrouped according to the linkage learned when the optimization procedure is considered as stagnant. Experiments are conduct on CEC'10 LSGO benchmark function. The test results confirm the validity of this method.

Cooperative co-evolutionary algorithm (CC) which runs in a divide-and-conquer manner is effective to solve large-scale global optimization (LSGO) problems. Multi-modal optimization (MMO) intends to locate multiple optimal solutions. Using MMO methods in CC algorithm would be beneficial, because MMO optimizer can provide more information about the landscapes. In this paper, a bi-objective selection is proposed to introduce imbalance among the subpopulations of a MMO optimizer. Only the highly representative subpopulations will be active and evolved in the MMO procedure. With this imbalanced MMO technique, the CC's subcomponents could obtain sufficient coevolutionary information (multiple optima) from each other. In addition, more computational resources could be saved and used in CC procedure. Experiments and statistical comparisons are conducted on LSGO benchmark functions to verify the effectiveness of the proposed method. The results indicate that the proposed algorithm significantly outperforms seven state-of-the-art large-scale CC algorithms.

Cooperative coevolutionary algorithms (CCEAs) divides a problem into several components and optimizes them independently. Some coevolutionary information will be lost due to the search space separation. This may lead some algorithmic pathologies, such as relative overgeneralization. In addition, according to the interactive nature of the CCEA, the coevolutionary landscapes are dynamic. In this paper, a multipopulation strategy is proposed to simultaneously search local or global optima in each dynamic landscape and provide them to the other components. Besides, a grid-based archive scheme is proposed to archive these historic collaborators for reasonable fitness evaluation. Two benchmark problems were used to test and compare the proposed algorithm to three classical CCEAs. Experimental results show that the proposed algorithm effectively counteract relative overgeneralization pathology and significantly improve the rate of converging to global optimum.