A Pheromone-Inspired Monitoring Strategy Using a Swarm of Underwater Robots

Guannan Li, Chao Chen, Chao Geng, Meng Li, Hongli Xu, Yang Lin2019 · DOI

TL;DR

This paper introduces a novel pheromone-inspired strategy for underwater robot swarms to monitor large areas and search for static targets. By leveraging indirect communication through a network of underwater nodes and a virtual pheromone map, the swarm can coordinate effectively without direct, bandwidth-limited underwater communication, outperforming random search methods.

What Problem Does This Paper Address?

Monitoring vast underwater areas and searching for static targets (like shipwrecks or resources) is a critical but challenging task. Traditional methods using single robots are slow, and swarms of underwater robots face significant hurdles: direct communication is severely limited by low bandwidth (acoustic) or short range (optical), making coordination of large swarms over extensive areas impractical. This paper tackles the problem of coordinating a large underwater robot swarm for efficient, cooperative monitoring and target search without relying on direct, real-time communication between robots.

Methods

The core of the proposed solution is a pheromone-inspired indirect communication strategy. Instead of robots talking directly, they interact with a network of underwater communication nodes.

  • Communication Network: A set of fixed underwater communication nodes are deployed in the Area of Interest (AOI). Robots visit these nodes to upload their local observations and download an updated "pheromone map" representing the collective knowledge of the swarm.
  • Virtual Pheromone Map: This map acts as a shared environment where robots can "leave" and "read" information. It's represented as a matrix where each cell's value indicates the density of virtual pheromone. This density can represent explored areas or detected targets.
  • Pheromone-Inspired Controller: Each robot has a controller with two layers:
  • Pheromone Map Layer: Maintains and updates the robot's local view of the pheromone map.
  • Behavior Law Layer: Uses the pheromone map to decide the robot's next action (search, visit node, report).
  • Behavior Laws:
    • Local Behavior (flocal): Robots move towards the neighbor cell with the lowest pheromone density, mimicking ant foraging.
  • Global Behavior (fglobal): An improved strategy that considers pheromone information from the entire AOI, not just immediate neighbors, leading to more balanced exploration.
  • Dynamic Re-visit Time (tback): The paper also proposes a mechanism for robots to dynamically adjust their time between visiting communication nodes, optimizing performance by learning from past experiences.

Datasets

Not discussed in the paper. The authors rely on simulations and a real-world experiment using USVs.

Key Insights and Ideas

  • Indirect Communication is Key: For large underwater swarms, indirect communication via a node network is a viable alternative to direct communication, overcoming bandwidth and range limitations.
  • Virtual Pheromones for Coordination: Mimicking social insect behavior with virtual pheromones provides a robust and scalable way to coordinate swarm actions without explicit leader-robot or robot-robot communication.
  • Global vs. Local Information: Considering global pheromone information (fglobal) significantly improves exploration efficiency and coverage compared to relying solely on local information (flocal). This leads to more balanced swarm behavior and better target detection.
  • Dynamic tback for Optimization: Automatically adjusting the time robots spend between visiting communication nodes can improve performance, especially for the local strategy, by avoiding suboptimal fixed intervals.

Key Results

  • Simulation:
    • The fglobal strategy consistently outperformed flocal and random search in terms of coverage rate, regardless of swarm size or tback value.
  • Both fglobal and flocal significantly outperformed random search.
  • Increasing swarm size generally improved coverage rate for all strategies.
  • Dynamic adjustment of tback showed superior performance for flocal compared to fixed tback values, and acceptable performance for fglobal.
  • Real-World Experiment (USVs):
    • The fglobal strategy achieved 100% coverage rate with a mean square error of 0.2899.
  • The flocal strategy achieved 88% coverage rate with a mean square error of 0.3594.
  • These results validated the simulation findings, demonstrating the superiority of global pheromone information.

Limitations and Future Work

  • Computational Cost: The fglobal strategy, while effective, can require a high-performance computer, especially for large AOIs, increasing cost and energy consumption. The flocal strategy offers a trade-off with lower computational demands.
  • tback Optimization: While dynamic tback adjustment improved performance, it was based on individual robot history. Future work could explore using historical performance of other robots to further enhance this scheme.
  • Information Extraction: The current fglobal scheme primarily uses mean pheromone density. Future work aims to extract more valuable information from the global pheromone matrix to further improve swarm performance.
  • Real-World Underwater Deployment: The real-world experiment used USVs to mimic underwater robots. Future work will focus on demonstrating the strategy's effectiveness with actual underwater robots.
  • Predictive Capabilities: The authors plan to enhance the strategy by predicting global information based on past network data and demonstrating its effectiveness in real-world applications.

Summary updated

BibTeX
@misc{li2019pheromone,
  author = {Guannan Li and Chao Chen and Chao Geng and Meng Li and Hongli Xu and Yang Lin},
  title = {{A Pheromone-Inspired Monitoring Strategy Using a Swarm of Underwater Robots}},
  year = {2019},
  doi = {10.3390/s19194089},
}
Figure