A novel foraging algorithm for swarm robotics based on virtual pheromones and neural network

https://doi.org/10.1016/j.asoc.2020.106156Get rights and content

Highlights

  • A novel neural network based pheromone model of swarm foraging behavior.

  • An optimization method of key parameters for cooperative swarm foraging with mathematical modeling.

  • A cooperative foraging algorithm based on neural network and the optimization method.

Abstract

Swarm robotics is an emerging interdisciplinary field that has many potential real-world applications. Swarm robotics aims to produce robust, scalable, and flexible self-organizing behaviors through local interactions from a large number of simple robots. In this paper, a novel pheromone model of swarm foraging behavior is developed based on a neural network. The output of a single neuron corresponds to the density of a pheromone, which diffuses to neighboring neurons through their local connections. A neural network is updated based on the proposed evaporation model. Neural networks can often mimic the dynamics and features of pheromones. Therefore, in this work, we develop an optimization method to determine the key parameters of cooperative foraging based on mathematical modeling. The differential equation variables represent the number of foraging robots assigned different tasks. The solutions of the differential equations represent the dynamics of the foraging behavior. The key parameters that affect task allocation are determined to make optimal decision rules. Simulation experiments are conducted under different foraging scenarios. The experimental results demonstrate the effectiveness of the proposed pheromone model.

Introduction

Swarm intelligence is a discipline inspired by nature, especially social insects. This discipline focuses on the collective behavior of social swarms [1]. The desired swarm behavior emerges from a set of simple rules and the local interactions of a large number of homogeneous and simple individual robots. The self-organizing coordination mechanisms of social insects have been effectively implemented in swarm robotic systems [2], [3], [4]. The application of swarm intelligence in robotics is known as swarm robotics, which has been successfully applied to different fields, such as self-driving, delivery robots, autonomous agricultural robots and automated warehouses [reviewed in 5].

Swarm robotics are a relatively new approach to designing robust, scalable, and flexible collective behaviors in which swarm intelligence techniques are applied. Many different schemes for emerging collective behaviors have been proposed during recent years, such as fully distributed architecture (ALLIANCE) [6], parallel multi-agent architecture [7], task-oriented hierarchical control architecture [8], neural-endocrine architecture [9], cooperative architecture [10], and modular multi-agent architecture [11]. To date, a complete theoretical framework that constructs a control architecture for group behavior does not fully exist. In recent years, nature-inspired metaheuristic algorithms have been applied to swarm robotics applications. The most well-known swarm-based algorithms are: particle swarm optimization [12], ant colony optimization, bee algorithm [13], and fish-swarm algorithm [14]. The remarkable success of using social animal behavior to model emerging collective behaviors provides a best-known paradigm for swarm robotics. In addition, macroscopic modeling has been available to understand the individual robot characteristics and gain insight into the design and analysis of swarm robotics behaviors [15], [16], [17], [18]. Mathematical models are typically abstract versions of swarm robotics and are usually based on Markov processes. In practical implementations, thoughtful work must be performed to overcome the limitations of the models.

Although some emerging behavior models have been implemented in some swarm robotics platforms, the dynamic real-time control of a swarm robotic system is still considered challenging. Swarm foraging behavior is a classical benchmark problem in swarm robotics. The biggest challenge is to develop a self-organizing search and collection algorithm for swarm robotic foraging behaviors. This work targets two main contributions: to propose a novel neural network-based pheromone model of swarm foraging behavior and to study the optimization method for key parameters of cooperative swarm foraging with mathematical modeling. In this paper, a novel pheromone model of swarm foraging behavior is developed based on a neural network. If the foraging robots secretes pheromones, a corresponding neuron output will increase. The output will diffuse to neighboring neurons through local connections. Therefore, the output of the neural network will be updated based on the proposed pheromone evaporation model. The evolution of the neural network can mimic the features of pheromones. A parameter optimization algorithm, based on differential equations, is proposed in this work. The key variables of a set of rate equations are indexed by the number of foraging robots assigned to different tasks. Theoretical analysis shows that the waiting time of scouting robots to deliver food items is a critical parameter that affect the cooperation rate. We determine the optimal parameters of cooperative foraging to make the optimal control strategy. In addition, we present various scenarios that verify the effectiveness of the proposed pheromone model for cooperative swarm foraging.

The remainder of this work is organized as follows. The most relevant literature describing swarm foraging robotics are reviewed in Section 2. Section 3 proposes a novel foraging algorithm and an optimization method to find the cooperation rate. Section 4 describes the simulation experiments run for different foraging scenarios, lists the performance metrics, and begins the analysis of the results. The results are discussed and analyzed further in Section 5. Finally, some concluding remarks and future work are provided in Section 6.

Section snippets

Related works

Swarm robotics systems coordinate the behaviors of large numbers of simple mobile robots to emerge desired collective behaviors. The robots must avoid collisions and perform a set of tasks based on constraints of system. Therefore, communication and cooperation are the most challenging problems when training swarm robotics.

The proposed methods

In this section, we propose a foraging algorithm based on a neural network. Section 3.1 depicts the swarm mission. Section 3.2 addresses the proposed pheromone model based on a neural network. Section 3.3 describes the optimization method that determines cooperation rate.

Implementation

Two computer experiments were carried out with one and two food sources in a dynamic environment. The grid world is made up of 100 × 100 square lattices. Each grid item is sized within its assigned area (a square 5 cm on a side). The output of the neurons at the boundaries of the domain are set to a negative constant. These boundaries are similar to obstacles in the area, e.g., areas with repellent pheromones. The maximum simulation time is 200 s in each foraging experiment. The size of the

Discussion

We have constructed and analyzed a novel pheromone-based cooperative foraging behavior model based on the wave expansion of neural networks. A neural network has a single layer that corresponds to the discretized grid environment. Each neuron is a label that identifies a unique position in the work space. When an ant lays down a pheromone at a specific position, the corresponding neuron is activated by an external input. Neural activities in the field propagate in waves from the neuron, where

Conclusions and future works

In this paper a novel pheromone model of swarm foraging behavior is developed based on a neural network. A dynamic wave expansion neural network (DWENN) is used to model pheromone diffusion. The neurons of the neural network correspond to different positions in a pre-defined workspace. When the robots release pheromones into their environment, the corresponding neurons (units) will receive an external input. The pheromones will diffuse through the local connections between neurons. The

Declaration of Competing Interest

No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work. For full disclosure statements refer to https://doi.org/10.1016/j.asoc.2020.106156.

Acknowledgments

This work is supported by the National Natural Science Foundation of China under Grants (61573213, 61973184, 61673245, 61603214, 61803227), National Key Research and Development Plan of China under Grant 2017YFB1300205, Shandong Province Key Research and Development Plan, China under Grants (2016ZDJS02A07, 2018GGX101039), China Postdoctoral Science Foundation under Grant 2018M630778 and Independent Innovation Foundation of Shandong University under Grant 2018ZQXM005, and The Development Plan of

References (52)

  • PitonakovaL. et al.

    The information-cost-reward framework for understanding robot swarm foraging

    Swarm Intell.

    (2018)
  • ParkerL.E.

    ALLIANCE: An architecture for fault tolerant multirobot cooperation

    IEEE Trans. Robot. Autom.

    (1998)
  • SilvaI.R.M. et al.

    MO-MAHM: A parallel multi-agent architecture for hybridization of metaheuristics for multi-objective problems

  • LengY.Q. et al.

    Task-oriented hierarchical control architecture for swarm robotic system

    Nat. Comput.

    (2017)
  • TimmisJ. et al.

    A neural-endocrine architecture for foraging in swarm robotic systems

  • PeresJ. et al.

    A multi-agent architecture for swarm robotics systems

  • GbengaD.E. et al.

    Understanding the limitations of particle swarm algorithm for dynamic optimization tasks: A survey towards the singularity of PSO for swarm robotic applications

    ACM Comput. Surv.

    (2016)
  • H. Verlekar, K. Joshi, Ant & bee inspired foraging swarm robots using computer vision, in: Proceedings of the...
  • ShenY. et al.

    Energy-saving task assignment for robotic fish sensor network based on artificial fish swarm algorithm

  • Taylor-KingJ.P. et al.

    Mathematical modelling of turning delays in swarm robotics

    IMA J. Appl. Math.

    (2015)
  • AznarF. et al.

    A macroscopic model for high intensity radiofrequency signal detection in swarm robotics systems

    Int. J. Comput. Math.

    (2014)
  • LermanK. et al.

    A review of probabilistic macroscopic models for swarm robotic systems

  • LermanK. et al.

    Analysis of dynamic task allocation in multi-robot systems

    Int. J. Robot. Res.

    (2006)
  • PangB. et al.

    Self-organized task allocation in swarm robotics foraging based on dynamical response threshold approach

  • BrutschyA. et al.

    Self-organized task allocation to sequentially interdependent tasks in swarm robotics

    Auton. Agents Multi-Agent Syst.

    (2014)
  • NedjahN. et al.

    PSO-Based Distributed Algorithm for Dynamic Task Allocation in a Robotic Swarm

    (2015)
  • Cited by (13)

    • Cooperation of unmanned systems for agricultural applications: A theoretical framework

      2022, Biosystems Engineering
      Citation Excerpt :

      A centralised control strategy is typically considered in this case (Arguenon, Bergues-Lagarde, Rosenberger, Bro, & Smari, 2006; Chevalier, Copot, De Keyser, Hernandez, & Ionescu, 2015), but decentralised solutions may also be considered. swarm robots: characterised by a set of autonomous machines with simpler control strategies than other categories, and by a high interaction capability among players (Song et al., 2020). A relevant aspect of this strategy is that the desired system behaviour emerges only by considering the whole system, while single machines cannot usually reach the mission target alone (Brambilla, Ferrante, Birattari, & Dorigo, 2013).

    View all citing articles on Scopus
    View full text